import AspectRatioType from '../../../common/enums/AspectRatioType.enum'; import EventBus from '../EventBus'; import Logger from '../Logger'; import Settings from '../Settings'; import VideoData from '../video-data/VideoData'; import { Corner } from './enums/corner.enum'; import { VideoPlaybackState } from './enums/video-playback-state.enum'; import { GlCanvas } from './gl/GlCanvas'; import { AardCanvasStore } from './interfaces/aard-canvas-store.interface'; import { AardDetectionSample, generateSampleArray, resetSamples } from './interfaces/aard-detection-sample.interface'; import { AardStatus, initAardStatus } from './interfaces/aard-status.interface'; import { AardTestResults, initAardTestResults, resetAardTestResults } from './interfaces/aard-test-results.interface'; import { AardTimers, initAardTimers } from './interfaces/aard-timers.interface'; /** * /\ * //\\ Automatic * // \\ Aspect * // \\ Ratio * \\ Detector * //XXXX \\ * // \\ (Totes not a Witcher reference) * // \\ (Witcher 2 best Witcher) * //XXXXXXXXXXXXXX\\ * * How it works: */ /** * [ ] Draw frame to canvas * | * | * | 0. A few things about the frame. * | We imagine that the frame is divided in three regions. * | Edge regions span from left and right edges towards the * | center of the frame. The border between edge region and * | center region is user-defined. We trust center region * | more than we trust the edge regions, because edge regions * | may contain a watermark that we don't give a single fuck * | about. To make things simple, we'll assume that each video * | gets at most one (1) watermark, which appears in at most * | one (1) of the four edge panels. * | * | EDGE REGION % CENTER REGION % EDGE REGION * | ↓ ↓ * | +----------------+---------------------+----------------+ * | | : : | * | | : : | * | | : v i d e o : | * | | . . . . . . . . . . . . . . . . . . . . . . . . . . . | * | | : f r a m e : | * | | : : | * | | : : | * | +----------------+---------------------+----------------+ * | * | * | * | 1. Do some fast tests in order to determine whether * | we need to run the more expensive tests. * | (admittedly, letterbox shrink and grow tests might * | be more expensive than full edge detection, but * | _probably_ not due to data locality) * | * | * V πŸ”Ά Check pixels in the corner. Are they black? * < > β€”β€”β€”β€” no β€”β€”β€”β€”+ * | V * yes no black bars, nothing to be done β€”β€”β€”> πŸ›‘ END * | * V πŸ”Ά Did we detect black bars yet? * < > β€”β€”β€”β€” yes β€”β€”β€”+ * | | Check pixels at the inner edge of the * | ↓ πŸ”Ά black bar. Are there any non-black pixels? * no < >β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€” no β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”>β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”+ * | | | * | yes | * | | | * | | Are those non-black pixels near the | * V V πŸ”Ά one of the edges & within tolerance? V * | < >β€”β€” yes, only at one edge β€”β€”β€”β€”β€”>β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”+ * | | | * | no, there's too many non-black pixels / | * | non-black pixels are in the center | * | | | * +β€”<β€”β€”β€” Aspect ratio definitely changed | * | | * | Check pixels at the outer edges of the | * | area of the frame we think should con- | * V tain the image. Are at least ??% of | * | those pixels non-black? πŸ”ΆV * | no β€”β€”β€”β€”β€”β€”< > * | | | * +β€”<β€”β€”β€”β€”β€”β€”β€”β€”β€”<β€”β€”β€”β€”β€”β€”β€”β€” Aspect ratio probably changed yes * | | * | V * | Aspect ratio is same as before * | | * | 2. If we came this far, we need to run aspect | * | ratio detection. | * | | * | | * | | * V | * SCAN FRAME, COLUMN BY COLUMN | * FOR EACH COLUMN: V | * : | : | * : Check every pixel from the edge towards : | * : the center of the frame. Continue until : | * : reaching the first non-black pixel. : | * : | : | * : | Did we reach a non-black pixel : | * : V πŸ”Άbefore reaching center of frame? : V * : no β€”β€”β€”< >β€”β€”β€” yes : | * : V | : | * : Mark column as V : | * : invalid Check pixels to the left and right : | * : of detected pixel within certain : | * : distance : | * : | : | * : V πŸ”Ά Are all pixels non-black? : V * : yes β€”β€”β€”< >β€”β€”β€” no : | * : | V : | * : | Mark column as invalid. : | * : V : | * : Check pixels to the left and right of detected : | * : pixel within certain distance, but in the last : | * : row without detection : | * : | : | * : V πŸ”Ά Are all pixels black? ...... V * : yes β€”β€”β€”< >β€”β€”β€” no : | * : V | : | * : | V : | * : | Mark column as invalid : | * : | : | * : | Is pixel significantly brighter : | * : V πŸ”Ά than the black level? : V * : yes β€”β€” < > β€”β€”β€” no : | * : | | : | * : | Run gradient detection. : | * : | | : | * : | V πŸ”Ά Is gradient detected? : V * : | no β€”β€”β€” < > β€”β€”β€” yes : | * : V V | : | * : Record the detected V : | * : position Mark column as invalid : | * - - - - | - - - - - - - - - - - - - | * V | * Process image positions we detected, separately | * for upper and lower bars. | * Topmost detection is considered the only valid | * option (unless in edge section under certain | * circumstances). If that position appears more | * than once, that increases our confidence. | * | | * V πŸ”Ά Are we confident enough? V * yes β€”β€”β€” < > β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€”β€” no | * V | | * Aspect ratio is certain Aspect ratio not certain | * | | | * | | | * | | | * | | | * : : | * | * 2. Sometimes, there might be subtitles hardcoded in | * the black bars. If user enables this feature, then V * presence of subtitle should invalidate | * TODO: IMPLEMENT THIS | * | * : : | * | | | * | | | * V πŸ”Ά Is subtitle detection enabled? πŸ”Ά V \/ * < > β€”β€”β€”β€”β€”β€”β€” no no β€”β€”β€”β€”β€”β€” < > Detect | * | V V | Sub? πŸ”Ά V * | Apply aspect Do nothing <β€”]|[β€”β€”β€”β€” no β€”β€”β€” < > * | ratio | | | * yes +β€”β€”β€”> πŸ›‘ END <β€”β€”β€”+ yes yes * | | | * | | | * V V V * RUN SUBTITLE DETECTION RUN SUBTITLE DETECTION * V V V * - - - - - - - - - - - - - - - - - - - - - * : [ ] : * : | : * : V : * : Draw only the letterbox region of the video to : * : a separate canvas and check a few rows. : * : | : * : | Are there any non-black pixels : * : V πŸ”Ά in sampled rows of letterbox? : * : < > β€”β€”β€”β€” no β€”β€”β€”β€”β€”β€”+ : * : | | Were we certain about aspect : * : yes V πŸ”Ά ratio before checking for subs? : * : | < > β€”β€”β€”β€” no β€”β€”β€”β€”> Do nothing : * : | | V : * : | yes β€”β€”β€”> Apply or keep -+-> πŸ›‘ END : * : | aspect ratio : * : | : * : +β€”β€”β€”β€”β€”β€”β€”+ Were we confident enough about : * : V πŸ”Ά detected aspect ratio : * : no β€”β€”β€” < > β€”β€”β€” yes : * : | | : * : | Stop automatic aspect ratio detection : * : V V : * : Reset aspect ratio to original : * : | : * : V : * : πŸ›‘ END : * - - - - - - - - - - - - - - - - - - - - - * */ export class Aard { //#region configuration parameters private logger: Logger; private videoData: VideoData; private settings: Settings; private eventBus: EventBus; private arid: string; private eventBusCommands = { // 'get-aard-timing': [{ // function: () => this.handlePerformanceDataRequest() // }] }; //#endregion private video: HTMLVideoElement; private animationFrame: number; //#region internal state public status: AardStatus = initAardStatus(); private timers: AardTimers = initAardTimers(); private canvasStore: AardCanvasStore; private testResults: AardTestResults; private canvasSamples: AardDetectionSample; //#endregion //#region getters get defaultAr() { if (!this.video) { return undefined; } const ratio = this.video.videoWidth / this.video.videoHeight; if (isNaN(ratio)) { return undefined; } return ratio; } //#endregion getters //#region lifecycle constructor(videoData: VideoData){ this.logger = videoData.logger; this.videoData = videoData; this.video = videoData.video; this.settings = videoData.settings; this.eventBus = videoData.eventBus; this.initEventBus(); this.arid = (Math.random()*100).toFixed(); // we can tick manually, for debugging this.logger.log('info', 'init', `[ArDetector::ctor] creating new ArDetector. arid: ${this.arid}`); this.init(); } private initEventBus() { for (const action in this.eventBusCommands) { for (const command of this.eventBusCommands[action]) { this.eventBus.subscribe(action, command); } } } /** * Initializes Aard with default values and starts autodetection loop. * This method should only ever be called from constructor. */ private init() { this.canvasStore = { main: new GlCanvas(new GlCanvas({...this.settings.active.arDetect.canvasDimensions.sampleCanvas, id: 'main-gl'})), }; this.canvasSamples = { top: generateSampleArray( this.settings.active.arDetect.sampling.staticCols, this.settings.active.arDetect.canvasDimensions.sampleCanvas.width ), bottom: generateSampleArray( this.settings.active.arDetect.sampling.staticCols, this.settings.active.arDetect.canvasDimensions.sampleCanvas.width ), }; this.start(); } //#endregion /** * Starts autodetection loop. */ start() { if (this.videoData.resizer.lastAr.type === AspectRatioType.AutomaticUpdate) { // ensure first autodetection will run in any case this.videoData.resizer.lastAr = {type: AspectRatioType.AutomaticUpdate, ratio: this.defaultAr}; } // do full reset of test samples this.testResults = initAardTestResults(this.settings.active.arDetect); if (this.animationFrame) { window.cancelAnimationFrame(this.animationFrame); } this.status.aardActive = true; this.animationFrame = window.requestAnimationFrame( (ts: DOMHighResTimeStamp) => this.onAnimationFrame(ts)); } /** * Runs autodetection ONCE. * If autodetection loop is running, this will also stop autodetection loop. */ step() { this.stop(); this.main(); } /** * Stops autodetection. */ stop() { if (this.animationFrame) { window.cancelAnimationFrame(this.animationFrame); } } //#region animationFrame, scheduling, and other shit /** * Checks whether conditions for granting a frame check are fulfilled * @returns */ private canTriggerFrameCheck() { // if (this._paused || this._halted || this._exited) { // return false; // } // if video was paused & we know that we already checked that frame, // we will not check it again. const videoState = this.getVideoPlaybackState(); if (videoState !== VideoPlaybackState.Playing) { if (this.status.lastVideoStatus === videoState) { return false; } } this.status.lastVideoStatus = videoState; if (Date.now() < this.timers.nextFrameCheckTime) { return false; } this.timers.nextFrameCheckTime = Date.now() + this.settings.active.arDetect.timers.playing; return true; } private onAnimationFrame(ts: DOMHighResTimeStamp) { if (this.canTriggerFrameCheck()) { resetAardTestResults(this.testResults); resetSamples(this.canvasSamples); this.main(); } else { } this.animationFrame = window.requestAnimationFrame( (ts: DOMHighResTimeStamp) => this.onAnimationFrame(ts)); } //#endregion /** * Main loop for scanning aspect ratio changes */ private async main() { try { // We abuse a do-while loop to eat our cake (get early returns) // and have it, too (if we return early, we still execute code // at the end of this function) do { const imageData = await new Promise( resolve => { this.canvasStore.main.drawVideoFrame(this.video); resolve(this.canvasStore.main.getImageData()); } ); // STEP 1: // Test if corners are black. If they're not, we can immediately quit the loop. this.getBlackLevelFast( imageData, 3, 1, this.settings.active.arDetect.canvasDimensions.sampleCanvas.width, this.settings.active.arDetect.canvasDimensions.sampleCanvas.height ); if (this.testResults.notLetterbox) { // TODO: reset aspect ratio to "AR not applied" this.testResults.lastStage = 1; break; } // STEP 2: // Check if previously detected aspect ratio is still gucci. If it is, then // we can quit the loop without applying any aspect ratios (unless subtitle // detection is enabled, in which case we still run the subtitle test) this.checkLetterboxShrink( imageData, this.settings.active.arDetect.canvasDimensions.sampleCanvas.width, this.settings.active.arDetect.canvasDimensions.sampleCanvas.height ); if (! this.testResults.guardLine.invalidated) { this.checkLetterboxGrow( imageData, this.settings.active.arDetect.canvasDimensions.sampleCanvas.width, this.settings.active.arDetect.canvasDimensions.sampleCanvas.height ); } // Both need to be checked if (! (this.testResults.imageLine.invalidated || this.testResults.guardLine.invalidated)) { // TODO: ensure no aspect ratio changes happen this.testResults.lastStage = 2; break; } // STEP 3: // If we are here, we must do full aspect ratio detection. // After aspectRatioCheck is finished, we know how wide the letterbox is. this.aspectRatioCheck( imageData, this.settings.active.arDetect.canvasDimensions.sampleCanvas.width, this.settings.active.arDetect.canvasDimensions.sampleCanvas.height ); } while (false); // TODO: subtitle check goes here. // Note that subtitle check should reset aspect ratio outright, regardless of what other tests revealed. // Also note that subtitle check should run on newest aspect ratio data, rather than lag one frame behind // But implementation details are something for future Tam to figure out // if detection is uncertain, we don't do anything at all if (this.testResults.aspectRatioUncertain) { return; } // TODO: emit debug values if debugging is enabled this.testResults.isFinished = true; // if edge width changed, emit update event. if (this.testResults.aspectRatioUpdated) { this.videoData.resizer.updateAr({ type: AspectRatioType.AutomaticUpdate, ratio: this.getAr(), offset: this.testResults.letterboxOffset }); } } catch (e) { console.warn('[Ultrawidify] Aspect ratio autodetection crashed for some reason.\n\nsome reason:', e); this.videoData.resizer.setAr({type: AspectRatioType.AutomaticUpdate, ratio: this.defaultAr}); } } private getVideoPlaybackState(): VideoPlaybackState { try { if (this.video.ended) { return VideoPlaybackState.Ended; } else if (this.video.paused) { return VideoPlaybackState.Paused; } else if (this.video.error) { return VideoPlaybackState.Error; } else { return VideoPlaybackState.Playing; } } catch (e) { this.logger.log('warn', 'debug', `[ArDetect::getVideoPlaybackState] There was an error while determining video playback state.`, e); return VideoPlaybackState.Error; } } //#region buffer tests /** * Get black level of a given frame. We sample black level on very few * positions β€” just the corners of the frame. If letterboxing or pillarboxing * exists, then pixels in the corners of the frame should be the blackest * it gets. * * Sampling pattern are four lines, each shooting from its respective corner. * Value of 'sample' parameter determines how many pixels along this line we * are going to sample. Offset means how many pixels of those four lines we * are going to skip before we start sampling. * * xβ†’ 0 1 ... ... x-1 * y↓ Γ— ------------... ...------------ Γ— * 0 | 1 1 | * 1 | 2 2 | * : | . . : * : . . * * : . . : * | . . | * | 2 2 | * h-1 | 1 1 | * Γ— ------------... ...------------ Γ— * * * IMPORTANT NOTES * <> imageData is one-dimensional array, so we need to account for that. * <> blackLevel is the darkest brightest subpixel detected * <> If image has no crop, then this function WILL NOT get the true black level. * In that case, we don't get an accurate black level, but we know straight * away that the image is uncropped. If image is uncropped, we can skip other, * more expensive tests. * * @param imageData array of pixels (4 bytes/fields per pixel) * @param samples number of samples per corner * @param width width of the frame * @param height height of the frame */ private getBlackLevelFast(imageData: Uint8Array, samples: number, offset: number, width: number, height: number) { // there's 4 points for each sample, and 3 components for each of the sampling points. const pixelValues = new Array(samples * 12); let pvi = 0; /** * We should ensure we are accessing pixels in ordered manner in order to * take advantage of data locality. */ const end = offset + samples; for (let i = offset; i < end; i++) { const px_r = (i * width * 4) + (i * 4); // red component starts here pixelValues[pvi++] = imageData[px_r]; pixelValues[pvi++] = imageData[px_r + 1]; pixelValues[pvi++] = imageData[px_r + 2]; const endpx_r = px_r + (width * 4) - (i * 8) - 4; // -4 because 4 bytes per pixel, and - twice the offset to mirror the diagonal pixelValues[pvi++] = imageData[endpx_r]; pixelValues[pvi++] = imageData[endpx_r + 1]; pixelValues[pvi++] = imageData[endpx_r + 2]; } // now let's populate the bottom two corners for (let i = end; i --> offset;) { const row = height - i - 1; // since first row is 0, last row is height - 1 const px_r = (row * width * 4) + (i * 4); pixelValues[pvi++] = imageData[px_r]; pixelValues[pvi++] = imageData[px_r + 1]; pixelValues[pvi++] = imageData[px_r + 2]; const endpx_r = px_r + (width * 4) - (i * 8) - 4; // -4 because 4 bytes per pixel, and - twice the offset to mirror the diagonal pixelValues[pvi++] = imageData[endpx_r]; pixelValues[pvi++] = imageData[endpx_r + 1]; pixelValues[pvi++] = imageData[endpx_r + 2]; } let min = 255; let avg = 0; let p = 0; for (let i = 0; i < pixelValues.length; i++) { p = pixelValues[i]; i++; if (p < pixelValues[i]) { p = pixelValues[i]; } i++; if (p < pixelValues[i]) { p = pixelValues[i]; } avg += p; if (p < min) { min = p; } } // Avg only contains highest subpixel, // but there's 4 subpixels per sample. avg = avg / (samples * 4); // TODO: unhardcode these values this.testResults.notLetterbox = avg > (this.testResults.blackLevel); // only update black level if not letterbox. // NOTE: but maybe we could, if blackLevel can only get lower than // the default value. if (this.testResults.notLetterbox) { if (min < this.testResults.blackLevel) { this.testResults.blackLevel = min; this.testResults.blackThreshold = min + 16; } } } /** * Checks if letterbox has shrunk. If letterbox has shrunk (image portion of the frame grows), we invalidate * guard line data. Note that this function only sets testResults.guardline.invalidated=true, but does not * override current guardline values. * NOTE: if guardLine is invalidated, the function will also helpfully invalidate imageLine results. This * will happen because invalid blackLine logically implies invalid imageLine. * @param imageData * @param width * @param height */ private checkLetterboxShrink(imageData: Uint8Array, width: number, height: number) { // can't check guardline if guardline is not set up (correctly) if ( this.testResults.guardLine.top < 0 || this.testResults.guardLine.top > height || this.testResults.guardLine.bottom < 0 || this.testResults.guardLine.bottom > height ) { // we also need to reset guardline if out-of-bounds was detected, // otherwise edgeScan might not work correctly this.testResults.guardLine.top = -1; this.testResults.guardLine.bottom = -1; this.testResults.guardLine.invalidated = true; return; } let edgePosition = this.settings.active.arDetect.sampling.edgePosition; const segmentPixels = width * edgePosition; const edgeSegmentSize = segmentPixels * 4; // check the top { // no use in doing guardline tests if guardline hasn't been measured yet, or if // guardline is not defined. const rowStart = this.testResults.guardLine.top * width * 4; const firstSegment = rowStart + edgeSegmentSize; const rowEnd = rowStart + (width * 4) - 4; const secondSegment = rowEnd - edgeSegmentSize; let i = rowStart; while (i < firstSegment) { if ( imageData[i] > this.testResults.blackThreshold || imageData[i + 1] > this.testResults.blackThreshold || imageData[i + 2] > this.testResults.blackThreshold ) { this.testResults.guardLine.cornerPixelsViolated[Corner.TopLeft]++; } i += 4; } while (i < secondSegment) { if ( imageData[i] > this.testResults.blackThreshold || imageData[i + 1] > this.testResults.blackThreshold || imageData[i + 2] > this.testResults.blackThreshold ) { // DONT FORGET TO INVALIDATE GUARDL LINE this.testResults.guardLine.top = -1; this.testResults.guardLine.bottom = -1; this.testResults.guardLine.invalidated = true; return; }; i += 4; } while (i < rowEnd) { if ( imageData[i] > this.testResults.blackThreshold || imageData[i + 1] > this.testResults.blackThreshold || imageData[i + 2] > this.testResults.blackThreshold ) { this.testResults.guardLine.cornerPixelsViolated[Corner.TopRight]++; } i += 4; // skip over alpha channel } } // check bottom { const rowStart = this.testResults.guardLine.bottom * width * 4; const firstSegment = rowStart + edgeSegmentSize; const rowEnd = rowStart + (width * 4) - 4; const secondSegment = rowEnd - edgeSegmentSize; let i = rowStart; if (i % 4) { i += 4 - (i % 4); } while (i < firstSegment) { if ( imageData[i] > this.testResults.blackThreshold || imageData[i + 1] > this.testResults.blackThreshold || imageData[i + 2] > this.testResults.blackThreshold ) { this.testResults.guardLine.cornerPixelsViolated[Corner.BottomLeft]++; } i += 4; // skip over alpha channel } if (i % 4) { i += 4 - (i % 4); } while (i < secondSegment) { if ( imageData[i] > this.testResults.blackThreshold || imageData[i + 1] > this.testResults.blackThreshold || imageData[i + 2] > this.testResults.blackThreshold ) { // DONT FORGET TO INVALIDATE GUARDL LINE this.testResults.guardLine.top = -1; this.testResults.guardLine.bottom = -1; this.testResults.guardLine.invalidated = true; return; }; i += 4; } if (i % 4) { i += 4 - (i % 4); } while (i < rowEnd) { if ( imageData[i] > this.testResults.blackThreshold || imageData[i + 1] > this.testResults.blackThreshold || imageData[i + 2] > this.testResults.blackThreshold ) { this.testResults.guardLine.cornerPixelsViolated[Corner.BottomRight]++; } i += 4; // skip over alpha channel } } const maxViolations = segmentPixels * 0.20; // TODO: move the 0.2 threshold into settings console.log('Corner violations counts β€” segment px & max violations,', segmentPixels, maxViolations ) // we won't do a loop for this few elements // corners with stuff in them will also be skipped in image test this.testResults.guardLine.cornerViolated[0] = this.testResults.guardLine.cornerPixelsViolated[0] > maxViolations; this.testResults.guardLine.cornerViolated[1] = this.testResults.guardLine.cornerPixelsViolated[1] > maxViolations; this.testResults.guardLine.cornerViolated[2] = this.testResults.guardLine.cornerPixelsViolated[2] > maxViolations; this.testResults.guardLine.cornerViolated[3] = this.testResults.guardLine.cornerPixelsViolated[3] > maxViolations; const maxInvalidCorners = 1; // TODO: move this into settings β€” by default, we allow one corner to extend past the // guard line in order to prevent watermarks/logos from preventing cropping the video // this works because +true converts to 1 and +false converts to 0 const dirtyCount = +this.testResults.guardLine.cornerViolated[0] + +this.testResults.guardLine.cornerViolated[1] + +this.testResults.guardLine.cornerViolated[2] + +this.testResults.guardLine.cornerViolated[3]; if (dirtyCount > maxInvalidCorners) { this.testResults.guardLine.invalidated = true; this.testResults.imageLine.invalidated = true; } else { this.testResults.guardLine.invalidated = false; } } /** * Checks if letterbox has grown. This test is super-efficient on frames that aren't dark, * but is also rather inefficient if the frame is overly dark. Note that this function merely * sets testResults.imageLine.invalidated to `true`. Correcting actual values is done during * aspect ratio detection. * TODO: maybe consider checking fewer pixels per line * @param imageData * @param width * @param height */ private checkLetterboxGrow(imageData: Uint8Array, width: number, height: number) { if ( this.testResults.imageLine.top < 0 || this.testResults.imageLine.top > height || this.testResults.imageLine.bottom < 0 || this.testResults.imageLine.bottom > height ) { this.testResults.imageLine.invalidated = true; return; } let edgePosition = this.settings.active.arDetect.sampling.edgePosition; const segmentPixels = width * edgePosition; const edgeSegmentSize = segmentPixels * 4; const detectionThreshold = width * 0.1; // TODO: unhardcoide and put into settings. Is % of total width. let imagePixel = false; let pixelCount = 0; // check the top { const rowStart = this.testResults.imageLine.top * width * 4; const firstSegment = rowStart + edgeSegmentSize; const rowEnd = rowStart + (width * 4) - 4; const secondSegment = rowEnd - edgeSegmentSize; let i = rowStart; // we don't run image detection in corners that may contain logos, as such corners // may not be representative if (! this.testResults.guardLine.cornerViolated[Corner.TopLeft]) { while (i < firstSegment) { imagePixel = false; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; if (imagePixel && ++pixelCount > detectionThreshold) { return; }; i++; // skip over alpha channel } } while (i < secondSegment) { imagePixel = false; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; if (imagePixel && ++pixelCount > detectionThreshold) { return; }; i++; // skip over alpha channel } if (! this.testResults.guardLine.cornerViolated[Corner.TopRight]) { while (i < rowEnd) { imagePixel = false; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; if (imagePixel && ++pixelCount > detectionThreshold) { return; }; i++; // skip over alpha channel } } // we don't run image detection in corners that may contain logos, as such corners // may not be representative if (! this.testResults.guardLine.cornerViolated[Corner.TopLeft]) { while (i < firstSegment) { imagePixel = false; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; if (imagePixel && ++pixelCount > detectionThreshold) { return; }; i++; // skip over alpha channel } } while (i < secondSegment) { imagePixel = false; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; if (imagePixel && ++pixelCount > detectionThreshold) { return; }; i++; // skip over alpha channel } if (! this.testResults.guardLine.cornerViolated[Corner.TopRight]) { while (i < rowEnd) { imagePixel = false; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; if (imagePixel && ++pixelCount > detectionThreshold) { return; }; i++; // skip over alpha channel } } } // check the bottom { const rowStart = this.testResults.imageLine.bottom * width * 4; const firstSegment = rowStart + edgeSegmentSize; const rowEnd = rowStart + (width * 4) - 4; const secondSegment = rowEnd - edgeSegmentSize; let i = rowStart; // we don't run image detection in corners that may contain logos, as such corners // may not be representative if (! this.testResults.guardLine.cornerViolated[Corner.TopLeft]) { while (i < firstSegment) { imagePixel = false; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; if (imagePixel && ++pixelCount > detectionThreshold) { return; }; i++; // skip over alpha channel } } while (i < secondSegment) { imagePixel = false; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; if (imagePixel && ++pixelCount > detectionThreshold) { return; }; i++; // skip over alpha channel } if (! this.testResults.guardLine.cornerViolated[Corner.TopRight]) { while (i < rowEnd) { imagePixel = false; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; if (imagePixel && ++pixelCount > detectionThreshold) { return; }; i++; // skip over alpha channel } } // we don't run image detection in corners that may contain logos, as such corners // may not be representative if (! this.testResults.guardLine.cornerViolated[Corner.TopLeft]) { while (i < firstSegment) { imagePixel = false; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; if (imagePixel && ++pixelCount > detectionThreshold) { return; }; i++; // skip over alpha channel } } while (i < secondSegment) { imagePixel = false; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; if (imagePixel && ++pixelCount > detectionThreshold) { return; }; i++; // skip over alpha channel } if (! this.testResults.guardLine.cornerViolated[Corner.TopRight]) { while (i < rowEnd) { imagePixel = false; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; imagePixel ||= imageData[i++] > this.testResults.blackThreshold; if (imagePixel && ++pixelCount > detectionThreshold) { return; }; i++; // skip over alpha channel } } } // if we came this far, we didn't get enough non-black pixels in order // to detect image. imageLine needs to be invalidated. this.testResults.imageLine.invalidated = true; } /** * Tries to detect aspect ratio. * * β€”β€”β€”< FAQ >β€”β€”β€” * Why not binary search? * * - Binary search is prone to false detections in certain * scenarios where multiple horizontal dark and bright areas * are present in the frame, e.g. window blinds * * * P.S.: * Future Tam, don't fucking think about that. I did the homework, * you aren't getting paid enough to find a way to make binary * search work. Go and work on a neat mini or an ambitious cosplay, * Chrome Web Store absolutely does not deserve this level of effort, * If you wanna chase imaginary internet approval points, then cosplay * and minis ripped from GW2 and Styx require much less sanity and * provide much more imaginary reddit points. * * Also maybe finish that story you're writing since 2009 if you * haven't already. Or learn Godot. */ private aspectRatioCheck(imageData: Uint8Array, width: number, height: number) { // this costs us tiny bit of overhead, but it makes code slightly more // manageable. We'll be making this tradeoff here, mostly due to the // fact that it makes the 'if' statement governing gradient detection // bit more nicely visible (instead of hidden among spagheti) this.edgeScan(imageData, width, height); this.validateEdgeScan(imageData, width, height); // TODO: _if gradient detection is enabled, then: this.sampleForGradient(imageData, width, height); this.processScanResults(imageData, width, height); } /** * Detects positions where frame stops being black and begins to contain image. * @param imageData * @param width * @param height */ private edgeScan(imageData: Uint8Array, width: number, height: number) { const detectionLimit = this.settings.active.arDetect.edgeDetection.thresholds.edgeDetectionLimit; let mid = ~~(height / 2); let topStart = 0; let topEnd = mid; let bottomStart = height; let bottomEnd = mid; let rowOffset = 0; /** * We can use invalidated blackbar and imagebar data to make some inferences * about where to find our letterbox. This test is all the data we need to check * if valid guardLine has ever been set, since guardLine and imageLine are set * in tandem (either both exist, or neither does (-1)). */ if (this.testResults.guardLine.top > 0) { // if guardLine is invalidated, then the new edge of image frame must be // above former guardline. Otherwise, it's below it. if (this.testResults.guardLine.invalidated) { topEnd = this.testResults.guardLine.top; bottomEnd = this.testResults.guardLine.bottom; } else { topStart = this.testResults.imageLine.top; bottomStart = this.testResults.imageLine.bottom; } } let row: number, i: number, x: number, isImage: boolean, finishedRows: number; // Detect upper edge { row = Math.max(topStart, 0); x = 0; isImage = false; finishedRows = 0; while (row < topEnd) { i = 0; rowOffset = row * 4 * width; // test the entire row while (i < this.canvasSamples.top.length) { // read x offset for the row we're testing, after this `i` points to the // result location x = this.canvasSamples.top[i++]; // check for image, after we're done `x` points to alpha channel isImage = imageData[rowOffset + x] > this.testResults.blackLevel || imageData[rowOffset + x + 1] > this.testResults.blackLevel || imageData[rowOffset + x + 2] > this.testResults.blackLevel; if (!isImage) { // TODO: maybe some day mark this pixel as checked by writing to alpha channel i++; continue; } if (this.canvasSamples.top[i] === -1) { this.canvasSamples.top[i] = row; finishedRows++; } i++; } // quit test early if we can if (finishedRows >= detectionLimit) { break; } row++; } } // Detect lower edge // NOTE: this part of the frame is checked less efficiently, because testResults // array is not oriented in optimal way. It could be fixed but refer to the `P.S.` // section of this function's description. { row = bottomStart; i = 0; x = 0; isImage = false; finishedRows = 0; while (row --> bottomEnd) { i = 0; rowOffset = row * 4 * width; // test the entire row while (i < this.canvasSamples.bottom.length) { // read x offset for the row we're testing, after this `i` points to the // result location x = this.canvasSamples.bottom[i++]; // check for image, after we're done `x` points to alpha channel isImage = imageData[rowOffset + x] > this.testResults.blackLevel || imageData[rowOffset + x + 1] > this.testResults.blackLevel || imageData[rowOffset + x + 2] > this.testResults.blackLevel; if (!isImage) { // TODO: maybe some day mark this pixel as checked by writing to alpha channel i++; continue; } if (this.canvasSamples.bottom[i] === -1) { this.canvasSamples.bottom[i] = row; finishedRows++; } i++; } // quit test early if we can if (finishedRows >= detectionLimit) { break; } } } } /** * Validates edge scan results. * * We check _n_ pixels to the left and to the right of detection, one row above * the detection (or under, when checking the bottom letterbox). If there's anything * non-black in this area, we invalidate the detection by setting the relevant * `canvasSample` to -1. * * For bottom rows, this function also converts row to the offset from the bottom. * * Note that this function returns nothing β€” instead it modifies properties of this * class. We do this in order to reduce garbage generation. This code runs often, * therefore we prefer reusing variables to generating new ones whenever reasonably * possible (though not always). * * @param imageData * @param width * @param height */ private validateEdgeScan(imageData: Uint8Array, width: number, height: number) { let i = 0; let xs: number, xe: number, row: number; const slopeTestSample = this.settings.active.arDetect.edgeDetection.slopeTestWidth * 4; while (i < this.canvasSamples.top.length) { // calculate row offset: row = (this.canvasSamples.top[i + 1] - 1) * width * 4; xs = row + this.canvasSamples.top[i] - slopeTestSample; xe = row + this.canvasSamples.top[i] + slopeTestSample; while (xs < xe) { if ( imageData[xs] > this.testResults.blackThreshold || imageData[xs + 1] > this.testResults.blackThreshold || imageData[xs + 2] > this.testResults.blackThreshold ) { this.canvasSamples.top[i + 1] = -1; break; } xs += 4; } i += 2; } i = 0; let i1 = 0; while (i < this.canvasSamples.bottom.length) { // calculate row offset: i1 = i + 1; row = (this.canvasSamples.bottom[i1] - 1) * width * 4; xs = row + this.canvasSamples.bottom[i] - slopeTestSample; xe = row + this.canvasSamples.bottom[i] + slopeTestSample; while (xs < xe) { if ( imageData[xs] > this.testResults.blackThreshold || imageData[xs + 1] > this.testResults.blackThreshold || imageData[xs + 2] > this.testResults.blackThreshold ) { this.canvasSamples.bottom[i1] = -1; i += 2; break; } xs += 4; } if (this.canvasSamples.bottom[i1]) { this.canvasSamples.bottom[i1] = height - this.canvasSamples.bottom[i1]; } i += 2; } } /** * Tries to detect whether our detection is detecting a hard edge, or a gradient. * Gradients shouldn't count as detection. * @param imageData * @param width * @param height */ private sampleForGradient(imageData: Uint8Array, width: number, height: number) { let j = 0, maxSubpixel = 0, lastSubpixel = 0, firstSubpixel = 0, pixelOffset = 0; const sampleLimit = this.settings.active.arDetect.edgeDetection.gradientTestSamples; const blackThreshold = this.testResults.blackLevel + this.settings.active.arDetect.edgeDetection.gradientTestBlackThreshold; const realWidth = width * 4; upperEdgeCheck: for (let i = 1; i < this.canvasSamples.top.length; i += 2) { pixelOffset = this.canvasSamples.top[i] * realWidth + this.canvasSamples.top[i - 1] * 4; lastSubpixel = imageData[pixelOffset] > imageData[pixelOffset + 1] ? imageData[pixelOffset] : imageData[pixelOffset + 1]; lastSubpixel = lastSubpixel > imageData[pixelOffset + 1] ? lastSubpixel : imageData[pixelOffset]; firstSubpixel = lastSubpixel; // save it j = 1; while (j < sampleLimit) { maxSubpixel = imageData[pixelOffset] > imageData[pixelOffset + 1] ? imageData[pixelOffset] : imageData[pixelOffset + 1]; maxSubpixel = maxSubpixel > imageData[pixelOffset + 2] ? maxSubpixel : imageData[pixelOffset + 2]; /** * Some assumptions. * * * If max subpixel is above max threshold, we probs aren't in a gradient (as it would imply * too sudden of a change in pixel brightness) * * if we are looking at a gradient, then we expect every pixel to be brighter than the * previous one. If it isn't, then we probably aren't in a gradient. * * if delta is too big, we probably aren't looking at a gradient, either */ if ( maxSubpixel > blackThreshold || maxSubpixel < lastSubpixel || maxSubpixel - lastSubpixel > this.settings.active.arDetect.edgeDetection.gradientTestDeltaThreshold ) { continue upperEdgeCheck; } lastSubpixel = maxSubpixel; pixelOffset -= realWidth; j++; } // if we came this far, we're probably looking at a gradient β€” unless the last pixel of our sample // didn't change meaningfully from the first, in which chance we aren't. If the brightness increased // anywhere between 'not enough' and 'too much', we mark the measurement as invalid. if (lastSubpixel - firstSubpixel > this.settings.active.arDetect.edgeDetection.gradientTestMinDelta) { this.canvasSamples.top[i] = -1; } } lowerEdgeCheck: for (let i = 1; i < this.canvasSamples.bottom.length; i += 2) { pixelOffset = (height - this.canvasSamples.bottom[i]) * realWidth + this.canvasSamples.bottom[i - 1] * 4; lastSubpixel = imageData[pixelOffset] > imageData[pixelOffset + 1] ? imageData[pixelOffset] : imageData[pixelOffset + 1]; lastSubpixel = lastSubpixel > imageData[pixelOffset + 1] ? lastSubpixel : imageData[pixelOffset]; firstSubpixel = lastSubpixel; // save it j = 1; while (j < sampleLimit) { maxSubpixel = imageData[pixelOffset] > imageData[pixelOffset + 1] ? imageData[pixelOffset] : imageData[pixelOffset + 1]; maxSubpixel = maxSubpixel > imageData[pixelOffset + 2] ? maxSubpixel : imageData[pixelOffset + 2]; /** * Some assumptions. * * * If max subpixel is above max threshold, we probs aren't in a gradient (as it would imply * too sudden of a change in pixel brightness) * * if we are looking at a gradient, then we expect every pixel to be brighter than the * previous one. If it isn't, then we probably aren't in a gradient. * * if delta is too big, we probably aren't looking at a gradient, either */ if ( maxSubpixel > blackThreshold || maxSubpixel < lastSubpixel || maxSubpixel - lastSubpixel > this.settings.active.arDetect.edgeDetection.gradientTestDeltaThreshold ) { continue lowerEdgeCheck; } lastSubpixel = maxSubpixel; pixelOffset -= realWidth; j++; } // if we came this far, we're probably looking at a gradient β€” unless the last pixel of our sample // didn't change meaningfully from the first, in which chance we aren't. If the brightness increased // anywhere between 'not enough' and 'too much', we mark the measurement as invalid. if (lastSubpixel - firstSubpixel > this.settings.active.arDetect.edgeDetection.gradientTestMinDelta) { this.canvasSamples.bottom[i] = -1; } } } /** * Processes data gathered by edgeScan, validateEdgeScan, and sampleForGradient. * It takes samples and determines how wide the letterbox actually is. * @param imageData * @param width * @param height * @returns */ private processScanResults(imageData: Uint8Array, width: number, height: number) { /** * Few things to note β€” * our canvasSamples are positioned like this: * * |---0---1---2---3---| * 0 19 * * We need to figure out how many positions lie before and * after our cutoff mark (25% and 75% of width, respectively): * * |---0:--1---2--:3---| * | : : | * 0 5 15 19 * * In order to accurately determine whether column belongs * to edge region or not, we need to invent two extra imaginary * sampling position, in order to keep sampling position 0 at * 20% of the width. * * (NOTE: it was too late for me to actually think about whether this * holds any water, but it prolly doesn't matter too much anyway) */ const fullFence = this.settings.active.arDetect.sampling.staticCols + 1; const edgePosition = this.settings.active.arDetect.sampling.edgePosition; // remember: array has two places per sample position β€” hence x2 on the results const leftEdgeBoundary = ~~(fullFence * edgePosition) * 2; const rightEdgeBoundary = (this.settings.active.arDetect.sampling.staticCols - leftEdgeBoundary) * 2; let i: number; // Process top edge: i = 1; { // We'll just unroll this loop, too much overhead for 3 items this.testResults.aspectRatioCheck.topRows[0] = Infinity; this.testResults.aspectRatioCheck.topRows[1] = Infinity; this.testResults.aspectRatioCheck.topRows[2] = Infinity; this.testResults.aspectRatioCheck.topQuality[0] = 0; this.testResults.aspectRatioCheck.topQuality[1] = 0; this.testResults.aspectRatioCheck.topQuality[2] = 0; while (i < leftEdgeBoundary) { if (this.canvasSamples.top[i] > -1) { if (this.canvasSamples.top[i] <= this.testResults.aspectRatioCheck.topRows[0]) { this.testResults.aspectRatioCheck.topRows[0] = this.canvasSamples.top[i]; this.testResults.aspectRatioCheck.topQuality[0] = 0; } else if (this.canvasSamples.top[i] === this.testResults.aspectRatioCheck.topRows[0]) { this.testResults.aspectRatioCheck.topQuality[0]++; } } i += 2; } while (i < rightEdgeBoundary) { if (this.canvasSamples.top[i] > -1) { if (this.canvasSamples.top[i] <= this.testResults.aspectRatioCheck.topRows[1]) { this.testResults.aspectRatioCheck.topRows[1] = this.canvasSamples.top[i]; this.testResults.aspectRatioCheck.topQuality[1] = 0; } else if (this.canvasSamples.top[i] === this.testResults.aspectRatioCheck.topRows[1]) { this.testResults.aspectRatioCheck.topQuality[1]++; } } i += 2; } while (i < this.canvasSamples.top.length) { if (this.canvasSamples.top[i] > -1) { if (this.canvasSamples.top[i] <= this.testResults.aspectRatioCheck.topRows[2]) { this.testResults.aspectRatioCheck.topRows[2] = this.canvasSamples.top[i]; this.testResults.aspectRatioCheck.topQuality[2] = 0; } else if (this.canvasSamples.top[i] === this.testResults.aspectRatioCheck.topRows[2]) { this.testResults.aspectRatioCheck.topQuality[2]++; } } i += 2; } } // Process bottom edge i = 1; { // We'll just unroll this loop, too much overhead for 3 items this.testResults.aspectRatioCheck.bottomRows[0] = Infinity; this.testResults.aspectRatioCheck.bottomRows[1] = Infinity; this.testResults.aspectRatioCheck.bottomRows[2] = Infinity; this.testResults.aspectRatioCheck.bottomQuality[0] = 0; this.testResults.aspectRatioCheck.bottomQuality[1] = 0; this.testResults.aspectRatioCheck.bottomQuality[2] = 0; while (i < leftEdgeBoundary) { if (this.canvasSamples.bottom[i] > -1) { if (this.canvasSamples.bottom[i] <= this.testResults.aspectRatioCheck.bottomRows[0]) { this.testResults.aspectRatioCheck.bottomRows[0] = this.canvasSamples.bottom[i]; this.testResults.aspectRatioCheck.bottomQuality[0] = 0; } else if (this.canvasSamples.bottom[i] === this.testResults.aspectRatioCheck.bottomRows[0]) { this.testResults.aspectRatioCheck.bottomQuality[0]++; } } i += 2; } while (i < rightEdgeBoundary) { if (this.canvasSamples.bottom[i] > -1) { if (this.canvasSamples.bottom[i] <= this.testResults.aspectRatioCheck.bottomRows[1]) { this.testResults.aspectRatioCheck.bottomRows[1] = this.canvasSamples.bottom[i]; this.testResults.aspectRatioCheck.bottomQuality[1] = 0; } else if (this.canvasSamples.bottom[i] === this.testResults.aspectRatioCheck.bottomRows[1]) { this.testResults.aspectRatioCheck.bottomQuality[1]++; } } i += 2; } while (i < this.canvasSamples.bottom.length) { if (this.canvasSamples.bottom[i] > -1) { if (this.canvasSamples.bottom[i] <= this.testResults.aspectRatioCheck.bottomRows[2]) { this.testResults.aspectRatioCheck.bottomRows[2] = this.canvasSamples.bottom[i]; this.testResults.aspectRatioCheck.bottomQuality[2] = 0; } else if (this.canvasSamples.bottom[i] === this.testResults.aspectRatioCheck.bottomRows[2]) { this.testResults.aspectRatioCheck.bottomQuality[2]++; } } i += 2; } } /** * Determining our best edge candidate should, in theory, go * something like this: * * [ start ] * | * < > Are detections from all three sections on the same row * / \ * yes no β€”β€”β€”β€”> further testing needed * V | * valid candidate | * < > Are corner sections different? * / \ * yes no β€”β€”β€”β€”β€”+ * | | is center section closer * does any section | < > to the edge of the frame? * match with center? < > / \ * / \ no yes β€”β€”> center gets authority * yes no V * / | Center result is probably bad, regardless * Is center above | | of score. No logo + edge gets authority. * the mismatched | | * section? < > Topmost (closest-to-frame-edge) option wins, * / \ but detection quality is shit. * yes no * V \ * Not a logo. Center authority, * V * Center authority. * * * ... however ... * In practice: if there's too much mismatch, we just label detection * as inconclusive and do nothing. Not paid enough to figure out the * worst 5% of cases. */ // TOP: if ( this.testResults.aspectRatioCheck.topRows[0] === this.testResults.aspectRatioCheck.topRows[1] && this.testResults.aspectRatioCheck.topRows[0] === this.testResults.aspectRatioCheck.topRows[2] ) { // All three detections are the same this.testResults.aspectRatioCheck.topCandidate = this.testResults.aspectRatioCheck.topRows[0]; this.testResults.aspectRatioCheck.topCandidateQuality = this.testResults.aspectRatioCheck.topQuality[0] + this.testResults.aspectRatioCheck.topQuality[1] + this.testResults.aspectRatioCheck.topQuality[2]; } else if (this.testResults.aspectRatioCheck.topRows[0] === this.testResults.aspectRatioCheck.topRows[2]) { // Corners are the same, but different from center if (this.testResults.aspectRatioCheck.topRows[0] > this.testResults.aspectRatioCheck.topRows[1]) { // Corners are above center. this.testResults.aspectRatioCheck.topCandidate = this.testResults.aspectRatioCheck.topRows[0]; this.testResults.aspectRatioCheck.topCandidateQuality = this.testResults.aspectRatioCheck.topQuality[0] + this.testResults.aspectRatioCheck.topQuality[2] } else { // Corners are below center this.testResults.aspectRatioCheck.topCandidate = this.testResults.aspectRatioCheck.topRows[1]; this.testResults.aspectRatioCheck.topCandidateQuality = this.testResults.aspectRatioCheck.topQuality[1] } } else { // Corners are different. if ( this.testResults.aspectRatioCheck.topRows[0] !== this.testResults.aspectRatioCheck.topRows[1] && this.testResults.aspectRatioCheck.topRows[2] !== this.testResults.aspectRatioCheck.topRows[1] ) { // Center and matches neither of the corners. // TODO: maybe we can figure out to guess aspect ratio in scenarios like this. // But for the time being, just slap it with "inconclusive". this.testResults.aspectRatioUncertain = true; return; } else { // center matches one of the corners this.testResults.aspectRatioCheck.topCandidate = this.testResults.aspectRatioCheck.topRows[1]; this.testResults.aspectRatioCheck.topCandidateQuality = this.testResults.aspectRatioCheck.topQuality[1]; if (this.testResults.aspectRatioCheck.topRows[0] === this.testResults.aspectRatioCheck.topRows[1]) { this.testResults.aspectRatioCheck.topCandidateQuality += this.testResults.aspectRatioCheck.topRows[0]; } else { this.testResults.aspectRatioCheck.topCandidateQuality += this.testResults.aspectRatioCheck.topRows[2]; } } } // BOTTOM // Note that bottomRows candidates are measured from the top // Well have to invert our candidate after we're done if ( this.testResults.aspectRatioCheck.bottomRows[0] === this.testResults.aspectRatioCheck.bottomRows[1] && this.testResults.aspectRatioCheck.bottomRows[0] === this.testResults.aspectRatioCheck.bottomRows[2] ) { // All three detections are the same this.testResults.aspectRatioCheck.bottomCandidate = this.testResults.aspectRatioCheck.bottomRows[0]; this.testResults.aspectRatioCheck.bottomCandidateQuality = this.testResults.aspectRatioCheck.bottomQuality[0] + this.testResults.aspectRatioCheck.bottomQuality[1] + this.testResults.aspectRatioCheck.bottomQuality[2]; } else if (this.testResults.aspectRatioCheck.bottomRows[0] === this.testResults.aspectRatioCheck.bottomRows[2]) { // Corners are the same, but different from center if (this.testResults.aspectRatioCheck.bottomRows[0] > this.testResults.aspectRatioCheck.bottomRows[1]) { // Corners are above center. this.testResults.aspectRatioCheck.bottomCandidate = this.testResults.aspectRatioCheck.bottomRows[0]; this.testResults.aspectRatioCheck.bottomCandidateQuality = this.testResults.aspectRatioCheck.bottomQuality[0] + this.testResults.aspectRatioCheck.bottomQuality[2] } else { // Corners are below center this.testResults.aspectRatioCheck.bottomCandidate = this.testResults.aspectRatioCheck.bottomRows[1]; this.testResults.aspectRatioCheck.bottomCandidateQuality = this.testResults.aspectRatioCheck.bottomQuality[1] } } else { // Corners are different. if ( this.testResults.aspectRatioCheck.bottomRows[0] !== this.testResults.aspectRatioCheck.bottomRows[1] && this.testResults.aspectRatioCheck.bottomRows[2] !== this.testResults.aspectRatioCheck.bottomRows[1] ) { // Center and matches neither of the corners. // TODO: maybe we can figure out to guess aspect ratio in scenarios like this. // But for the time being, just slap it with "inconclusive". this.testResults.aspectRatioUncertain = true; return; } else { // center matches one of the corners this.testResults.aspectRatioCheck.bottomCandidate = this.testResults.aspectRatioCheck.bottomRows[1]; this.testResults.aspectRatioCheck.bottomCandidateQuality = this.testResults.aspectRatioCheck.bottomQuality[1]; if (this.testResults.aspectRatioCheck.bottomRows[0] === this.testResults.aspectRatioCheck.bottomRows[1]) { this.testResults.aspectRatioCheck.bottomCandidateQuality += this.testResults.aspectRatioCheck.bottomRows[0]; } else { this.testResults.aspectRatioCheck.bottomCandidateQuality += this.testResults.aspectRatioCheck.bottomRows[2]; } } } /** * Get final results. * Let candidateA hold better-quality candidate, and let the candidateB hold the lower-quality candidate. * candidateA must match or exceed minQualitySingleEdge and candidateB must match or exceed minQualitySecondEdge. */ let candidateA, candidateB; if (this.testResults.aspectRatioCheck.bottomCandidateQuality > this.testResults.aspectRatioCheck.topCandidateQuality) { candidateA = this.testResults.aspectRatioCheck.bottomCandidate; candidateB = this.testResults.aspectRatioCheck.topCandidate; } else { candidateA = this.testResults.aspectRatioCheck.topCandidate; candidateB = this.testResults.aspectRatioCheck.bottomCandidate; } if ( candidateA < this.settings.active.arDetect.edgeDetection.thresholds.minQualitySingleEdge || candidateB < this.settings.active.arDetect.edgeDetection.thresholds.minQualitySecondEdge ) { this.testResults.aspectRatioUncertain = true; return; } const maxOffset = ~~(height * this.settings.active.arDetect.edgeDetection.maxLetterboxOffset) const diff = this.testResults.aspectRatioCheck.topCandidate - this.testResults.aspectRatioCheck.bottomCandidate; const candidateAvg = ~~((this.testResults.aspectRatioCheck.topCandidate + this.testResults.aspectRatioCheck.bottomCandidate) / 2); if (diff > maxOffset) { this.testResults.aspectRatioUncertain = true; return; } if (maxOffset > 2) { this.testResults.imageLine.top = this.testResults.aspectRatioCheck.topCandidate === Infinity ? -1 : this.testResults.aspectRatioCheck.topCandidate; this.testResults.imageLine.bottom = this.testResults.aspectRatioCheck.bottomCandidate === Infinity ? -1 : this.testResults.aspectRatioCheck.bottomCandidate; this.testResults.guardLine.top = Math.max(this.testResults.imageLine.top - 2, 0); this.testResults.guardLine.bottom = Math.max(this.testResults.imageLine.bottom + 2, this.canvasStore.main.height - 1); } this.testResults.aspectRatioUncertain = false; this.testResults.letterboxWidth = candidateAvg; this.testResults.letterboxOffset = diff; this.testResults.aspectRatioUpdated = true; } /** * Calculates video's current aspect ratio based on data in testResults. * @returns */ private getAr() { const fileAr = this.video.videoWidth / this.video.videoHeight; const canvasAr = this.canvasStore.main.width / this.canvasStore.main.height; const compensatedWidth = fileAr === canvasAr ? this.canvasStore.main.width : this.canvasStore.main.width * fileAr; return compensatedWidth / (this.canvasStore.main.height - (this.testResults.letterboxWidth * 2)); } //#endregion }