应为要用 java 截图，而后传递给 opencv 解决，java 中截图是应用下边的代码：
Robot robot = new Robot();
BufferedImage screenCapture = robot.createScreenCapture(new Rectangle(0, 0, 1920, 1080));然而在我的电脑的全屏截图要 50ms 左右，但当我缩小截图区域后，耗时会有一个显著的成比例缩小，这让我想要看一下为什么有这样。
private synchronized BufferedImage[]
createCompatibleImage(Rectangle screenRect, boolean isHiDPI) {checkScreenCaptureAllowed();
checkValidRect(screenRect);
BufferedImage lowResolutionImage;
BufferedImage highResolutionImage;
DataBufferInt buffer;
WritableRaster raster;
BufferedImage[] imageArray;
// 因为调用截图 api 返回的像素为 int[]，所有这里确定 RGB 的值别离在那些位
if (screenCapCM == null) {
/*
* Fix for 4285201
* Create a DirectColorModel equivalent to the default RGB ColorModel,
* except with no Alpha component.
*/
screenCapCM = new DirectColorModel(24,
/* red mask */ 0x00FF0000,
/* green mask */ 0x0000FF00,
/* blue mask */ 0x000000FF);
}
int[] bandmasks = new int[3];
bandmasks[0] = screenCapCM.getRedMask();
bandmasks[1] = screenCapCM.getGreenMask();
bandmasks[2] = screenCapCM.getBlueMask();
// 感觉是避免反复截图，期待下一帧吧
Toolkit.getDefaultToolkit().sync();
//
GraphicsConfiguration gc = GraphicsEnvironment
.getLocalGraphicsEnvironment()
.getDefaultScreenDevice().
getDefaultConfiguration();
gc = SunGraphicsEnvironment.getGraphicsConfigurationAtPoint(gc, screenRect.getCenterX(), screenRect.getCenterY());
AffineTransform tx = gc.getDefaultTransform();
double uiScaleX = tx.getScaleX();
double uiScaleY = tx.getScaleY();
int[] pixels;
// 我电脑这里 uiScaleX 和 uiScaleY 都是 1.x，所以走的 else 分支
if (uiScaleX == 1 && uiScaleY == 1) {pixels = peer.getRGBPixels(screenRect);
buffer = new DataBufferInt(pixels, pixels.length);
bandmasks[0] = screenCapCM.getRedMask();
bandmasks[1] = screenCapCM.getGreenMask();
bandmasks[2] = screenCapCM.getBlueMask();
raster = Raster.createPackedRaster(buffer, screenRect.width,
screenRect.height, screenRect.width, bandmasks, null);
SunWritableRaster.makeTrackable(buffer);
highResolutionImage = new BufferedImage(screenCapCM, raster,
false, null);
imageArray = new BufferedImage[1];
imageArray[0] = highResolutionImage;
} else {
Rectangle scaledRect;
if (peer.useAbsoluteCoordinates()) {
scaledRect = toDeviceSpaceAbs(gc, screenRect.x,
screenRect.y, screenRect.width, screenRect.height);
} else {
scaledRect = toDeviceSpace(gc, screenRect.x,
screenRect.y, screenRect.width, screenRect.height);
}
// 这里调用本地办法截图，这里是一个耗时点
pixels = peer.getRGBPixels(scaledRect);
// 构建解析 pixels 中数据的对象，buffer = new DataBufferInt(pixels, pixels.length);
raster = Raster.createPackedRaster(buffer, scaledRect.width,
scaledRect.height, scaledRect.width, bandmasks, null);
SunWritableRaster.makeTrackable(buffer);
highResolutionImage = new BufferedImage(screenCapCM, raster,
false, null);
// 这里大略意思就是，依据高分辨率图像，生成低分辨率图像，然而 drawImage 办法也挺耗时的
lowResolutionImage = new BufferedImage(screenRect.width,
screenRect.height, highResolutionImage.getType());
Graphics2D g = lowResolutionImage.createGraphics();
g.setRenderingHint(RenderingHints.KEY_INTERPOLATION,
RenderingHints.VALUE_INTERPOLATION_BILINEAR);
g.setRenderingHint(RenderingHints.KEY_RENDERING,
RenderingHints.VALUE_RENDER_QUALITY);
g.setRenderingHint(RenderingHints.KEY_ANTIALIASING,
RenderingHints.VALUE_ANTIALIAS_ON);
g.drawImage(highResolutionImage, 0, 0,
screenRect.width, screenRect.height,
0, 0, scaledRect.width, scaledRect.height, null);
g.dispose();
if(!isHiDPI) {imageArray = new BufferedImage[1];
imageArray[0] = lowResolutionImage;
} else {imageArray = new BufferedImage[2];
imageArray[0] = lowResolutionImage;
imageArray[1] = highResolutionImage;
}
}
return imageArray;
} 我把 pixels = peer.getRGBPixels(scaledRect); 办法独自拿进去测试，速度是 26ms，也就是其余操作是优化的。
我最开始是查看 opencv 中是否有用一个 int 报错一个像素的 rgb 值的内容，然而如同并没有。所以我就尝试手动解析 pixels，从中提取 r，g，b 的值。
// 截图
pixels = peer.getRGBPixels(screenRect);
// 解析像素
int length = screenRect.width * screenRect.height;
byte[] imgBytes = new byte[length * 3];
int byteIndex = 0;
for (int i = 0, pixel = 0; i < length; i++) {pixel = pixels[i];
// pixel 中是依照 rgb 格局排序，然而 opencv 默认是 bgr 格局
imgBytes[byteIndex++] = (byte) (pixel);
pixel = pixel >> 8;
imgBytes[byteIndex++] = (byte) (pixel);
imgBytes[byteIndex++] = (byte) (pixel >> 8);
}通过测试，这里解析只须要 3~4ms，而后把这个 byte[] 传递给 mat 就好了。整体截一张图并传递给 opencv 的耗时为 30ms。
最初试了下在屏幕画面继续变动的状况下，一次截图并传给 opencv 的耗时为 35ms。