一、OpenCV在Android中的集成与配置

1.1 OpenCV简介

OpenCV（Open Source Computer Vision Library）是一个开源的计算机视觉和机器学习软件库，包含超过2500种优化算法，涵盖图像处理、模式识别、机器学习等众多领域。在Android平台上，OpenCV为开发者提供了强大的图像处理能力，可以用于实现各种复杂的计算机视觉应用。

1.2 在Android Studio中集成OpenCV

1.2.1 通过Gradle依赖集成

最简单的方式是通过Gradle添加OpenCV依赖：

dependencies {implementation 'org.opencv:opencv-android:4.5.5'
}

1.2.2 通过模块方式集成

更灵活的方式是下载OpenCV Android SDK并将其作为模块导入：

1. 从OpenCV官网下载Android SDK
2. 在Android Studio中选择File -> New -> Import Module
3. 选择OpenCV SDK中的java文件夹
4. 修改模块的build.gradle文件，确保与主项目兼容

1.2.3 初始化OpenCV

在应用启动时需要初始化OpenCV库：

public class MainActivity extends AppCompatActivity {private BaseLoaderCallback mLoaderCallback = new BaseLoaderCallback(this) {@Overridepublic void onManagerConnected(int status) {if (status == LoaderCallbackInterface.SUCCESS) {Log.i("OpenCV", "OpenCV loaded successfully");} else {super.onManagerConnected(status);}}};@Overrideprotected void onResume() {super.onResume();if (!OpenCVLoader.initDebug()) {OpenCVLoader.initAsync(OpenCVLoader.OPENCV_VERSION_3_0_0, this, mLoaderCallback);} else {mLoaderCallback.onManagerConnected(LoaderCallbackInterface.SUCCESS);}}
}

1.3 OpenCV基础类介绍

在Android中使用OpenCV主要涉及以下几个核心类：

1. Mat：OpenCV的基础矩阵类，用于存储图像数据
2. Bitmap：Android的位图类，需要与Mat相互转换
3. Imgproc：图像处理类，包含各种图像处理算法
4. Core：核心功能类，提供基本数学运算和矩阵操作
5. Feature2d：特征检测与描述类

二、指定区域图像抓取与对比

2.1 图像抓取基础

在Android中获取图像主要有以下几种方式：

1. 从相机捕获
2. 从图库选择
3. 从资源文件加载
4. 从网络下载

2.2 指定区域图像抓取实现

2.2.1 从Bitmap中截取指定区域

public static Bitmap cropBitmap(Bitmap srcBitmap, Rect region) {if (srcBitmap == null || region == null) return null;try {// 确保区域在图像范围内int x = Math.max(0, region.left);int y = Math.max(0, region.top);int width = Math.min(srcBitmap.getWidth() - x, region.width());int height = Math.min(srcBitmap.getHeight() - y, region.height());return Bitmap.createBitmap(srcBitmap, x, y, width, height);} catch (Exception e) {e.printStackTrace();return null;}
}

2.2.2 使用OpenCV截取指定区域

public static Mat cropImage(Mat srcMat, Rect roi) {if (srcMat.empty() || roi == null) return null;try {// 确保ROI在图像范围内Rect adjustedRoi = new Rect(Math.max(0, roi.x),Math.max(0, roi.y),Math.min(srcMat.cols() - roi.x, roi.width),Math.min(srcMat.rows() - roi.y, roi.height));return new Mat(srcMat, adjustedRoi);} catch (Exception e) {e.printStackTrace();return new Mat();}
}

2.3 图像对比技术

2.3.1 均方误差(MSE)对比

public static double compareImagesMSE(Mat img1, Mat img2) {if (img1.rows() != img2.rows() || img1.cols() != img2.cols()) {throw new IllegalArgumentException("Images must have same dimensions");}Mat diff = new Mat();Core.absdiff(img1, img2, diff);diff.convertTo(diff, CvType.CV_32F);diff = diff.mul(diff);Scalar mse = Core.mean(diff);return (mse.val[0] + mse.val[1] + mse.val[2]) / 3;
}

2.3.2 结构相似性(SSIM)对比

public static double compareImagesSSIM(Mat img1, Mat img2) {// 转换为灰度图像Mat gray1 = new Mat();Mat gray2 = new Mat();Imgproc.cvtColor(img1, gray1, Imgproc.COLOR_BGR2GRAY);Imgproc.cvtColor(img2, gray2, Imgproc.COLOR_BGR2GRAY);// 参数设置final double C1 = 6.5025, C2 = 58.5225;int d = CvType.CV_32F;Mat I1 = new Mat(), I2 = new Mat();gray1.convertTo(I1, d);gray2.convertTo(I2, d);Mat I1_2 = I1.mul(I1);Mat I2_2 = I2.mul(I2);Mat I1_I2 = I1.mul(I2);// 计算均值Mat mu1 = new Mat(), mu2 = new Mat();Imgproc.GaussianBlur(I1, mu1, new Size(11, 11), 1.5);Imgproc.GaussianBlur(I2, mu2, new Size(11, 11), 1.5);Mat mu1_2 = mu1.mul(mu1);Mat mu2_2 = mu2.mul(mu2);Mat mu1_mu2 = mu1.mul(mu2);// 计算方差Mat sigma1_2 = new Mat(), sigma2_2 = new Mat(), sigma12 = new Mat();Imgproc.GaussianBlur(I1_2, sigma1_2, new Size(11, 11), 1.5);Core.subtract(sigma1_2, mu1_2, sigma1_2);Imgproc.GaussianBlur(I2_2, sigma2_2, new Size(11, 11), 1.5);Core.subtract(sigma2_2, mu2_2, sigma2_2);Imgproc.GaussianBlur(I1_I2, sigma12, new Size(11, 11), 1.5);Core.subtract(sigma12, mu1_mu2, sigma12);// 计算SSIMMat t1 = new Mat(), t2 = new Mat(), t3 = new Mat();Core.multiply(mu1_mu2, new Scalar(2), t1);Core.add(t1, new Scalar(C1), t1);Core.multiply(sigma12, new Scalar(2), t2);Core.add(t2, new Scalar(C2), t2);Core.add(mu1_2, mu2_2, t3);Core.add(t3, new Scalar(C1), t3);Mat ssim_map = new Mat();Core.multiply(t1, t2, ssim_map);Core.divide(ssim_map, t3, ssim_map);Scalar mssim = Core.mean(ssim_map);return mssim.val[0];
}

2.3.3 特征点匹配对比

public static double compareImagesFeatureMatching(Mat img1, Mat img2) {// 转换为灰度图像Mat gray1 = new Mat();Mat gray2 = new Mat();Imgproc.cvtColor(img1, gray1, Imgproc.COLOR_BGR2GRAY);Imgproc.cvtColor(img2, gray2, Imgproc.COLOR_BGR2GRAY);// 检测ORB特征点ORB orb = ORB.create();MatOfKeyPoint keypoints1 = new MatOfKeyPoint();MatOfKeyPoint keypoints2 = new MatOfKeyPoint();Mat descriptors1 = new Mat();Mat descriptors2 = new Mat();orb.detectAndCompute(gray1, new Mat(), keypoints1, descriptors1);orb.detectAndCompute(gray2, new Mat(), keypoints2, descriptors2);// 使用BFMatcher进行匹配BFMatcher matcher = BFMatcher.create(BFMatcher.BRUTEFORCE_HAMMING);MatOfDMatch matches = new MatOfDMatch();matcher.match(descriptors1, descriptors2, matches);// 计算匹配质量List<DMatch> matchesList = matches.toList();double maxDist = 0;double minDist = 100;for (DMatch match : matchesList) {double dist = match.distance;if (dist < minDist) minDist = dist;if (dist > maxDist) maxDist = dist;}// 筛选好的匹配点List<DMatch> goodMatches = new ArrayList<>();for (DMatch match : matchesList) {if (match.distance <= Math.max(2 * minDist, 30.0)) {goodMatches.add(match);}}// 计算匹配率double matchRatio = (double)goodMatches.size() / matchesList.size();return matchRatio;
}

2.4 图像对比应用实例

2.4.1 图像相似度检测

public class ImageComparator {private static final double MSE_THRESHOLD = 1000;private static final double SSIM_THRESHOLD = 0.8;private static final double FEATURE_MATCH_THRESHOLD = 0.5;public enum ComparisonResult {VERY_SIMILAR,SIMILAR,DIFFERENT,INVALID}public static ComparisonResult compareImages(Bitmap bmp1, Bitmap bmp2, Rect roi1, Rect roi2) {try {// 转换Bitmap为MatMat mat1 = new Mat();Mat mat2 = new Mat();Utils.bitmapToMat(bmp1, mat1);Utils.bitmapToMat(bmp2, mat2);// 裁剪指定区域Mat cropped1 = cropImage(mat1, roi1);Mat cropped2 = cropImage(mat2, roi2);if (cropped1.empty() || cropped2.empty()) {return ComparisonResult.INVALID;}// 调整大小一致if (cropped1.size().width != cropped2.size().width || cropped1.size().height != cropped2.size().height) {Imgproc.resize(cropped2, cropped2, cropped1.size());}// 计算各种相似度指标double mse = compareImagesMSE(cropped1, cropped2);double ssim = compareImagesSSIM(cropped1, cropped2);double featureMatch = compareImagesFeatureMatching(cropped1, cropped2);// 综合判断if (mse < MSE_THRESHOLD && ssim > SSIM_THRESHOLD && featureMatch > FEATURE_MATCH_THRESHOLD) {return ComparisonResult.VERY_SIMILAR;} else if (ssim > SSIM_THRESHOLD || featureMatch > FEATURE_MATCH_THRESHOLD) {return ComparisonResult.SIMILAR;} else {return ComparisonResult.DIFFERENT;}} catch (Exception e) {e.printStackTrace();return ComparisonResult.INVALID;}}
}

2.4.2 图像差异可视化

public static Bitmap visualizeImageDifference(Bitmap bmp1, Bitmap bmp2) {try {// 转换Bitmap为MatMat mat1 = new Mat();Mat mat2 = new Mat();Utils.bitmapToMat(bmp1, mat1);Utils.bitmapToMat(bmp2, mat2);// 确保图像大小一致if (mat1.size().width != mat2.size().width || mat1.size().height != mat2.size().height) {Imgproc.resize(mat2, mat2, mat1.size());}// 计算差异Mat diff = new Mat();Core.absdiff(mat1, mat2, diff);// 增强差异可视化Core.normalize(diff, diff, 0, 255, Core.NORM_MINMAX);Imgproc.cvtColor(diff, diff, Imgproc.COLOR_BGR2RGB);// 转换回BitmapBitmap result = Bitmap.createBitmap(diff.cols(), diff.rows(), Bitmap.Config.ARGB_8888);Utils.matToBitmap(diff, result);return result;} catch (Exception e) {e.printStackTrace();return null;}
}

三、指定点颜色提取与对比

3.1 颜色空间基础

OpenCV支持多种颜色空间，常用的有：

1. RGB/BGR：默认颜色空间
2. HSV/HSL：色调、饱和度、亮度/明度
3. Lab：感知均匀的颜色空间
4. YCrCb：亮度和色度分量
5. Grayscale：灰度图像

3.2 指定点颜色提取

3.2.1 从Bitmap中获取像素颜色

public static int getPixelColor(Bitmap bitmap, int x, int y) {if (bitmap == null || x < 0 || y < 0 || x >= bitmap.getWidth() || y >= bitmap.getHeight()) {return 0;}return bitmap.getPixel(x, y);
}public static int[] getPixelColorARGB(Bitmap bitmap, int x, int y) {int pixel = getPixelColor(bitmap, x, y);return new int[] {(pixel >> 24) & 0xff, // Alpha(pixel >> 16) & 0xff, // Red(pixel >> 8) & 0xff,  // Greenpixel & 0xff          // Blue};
}

3.2.2 使用OpenCV获取像素颜色

public static double[] getPixelColor(Mat mat, int x, int y) {if (mat.empty() || x < 0 || y < 0 || x >= mat.cols() || y >= mat.rows()) {return new double[]{0, 0, 0};}double[] pixel = mat.get(y, x); // OpenCV中使用(row, col)顺序return pixel;
}public static Scalar getPixelColorScalar(Mat mat, int x, int y) {double[] pixel = getPixelColor(mat, x, y);if (pixel == null || pixel.length < 3) {return new Scalar(0, 0, 0);}return new Scalar(pixel[0], pixel[1], pixel[2]);
}

3.3 颜色空间转换

public static Mat convertColorSpace(Mat src, int conversionCode) {if (src.empty()) return new Mat();Mat dst = new Mat();try {Imgproc.cvtColor(src, dst, conversionCode);} catch (Exception e) {e.printStackTrace();}return dst;
}// 常用颜色空间转换代码
public static final int COLOR_BGR2RGB = Imgproc.COLOR_BGR2RGB;
public static final int COLOR_BGR2HSV = Imgproc.COLOR_BGR2HSV;
public static final int COLOR_BGR2Lab = Imgproc.COLOR_BGR2Lab;
public static final int COLOR_BGR2YCrCb = Imgproc.COLOR_BGR2YCrCb;
public static final int COLOR_BGR2GRAY = Imgproc.COLOR_BGR2GRAY;

3.4 颜色对比技术

3.4.1 欧氏距离颜色对比

public static double colorDistanceEuclidean(Scalar color1, Scalar color2) {double diffR = color1.val[0] - color2.val[0];double diffG = color1.val[1] - color2.val[1];double diffB = color1.val[2] - color2.val[2];return Math.sqrt(diffR * diffR + diffG * diffG + diffB * diffB);
}

3.4.2 CIEDE2000颜色差异算法

// 需要实现Lab颜色空间的Delta E计算
public static double colorDistanceCIEDE2000(Scalar lab1, Scalar lab2) {// 简化的Delta E 2000计算double L1 = lab1.val[0];double a1 = lab1.val[1];double b1 = lab1.val[2];double L2 = lab2.val[0];double a2 = lab2.val[1];double b2 = lab2.val[2];double deltaL = L2 - L1;double meanL = (L1 + L2) / 2;double C1 = Math.sqrt(a1 * a1 + b1 * b1);double C2 = Math.sqrt(a2 * a2 + b2 * b2);double meanC = (C1 + C2) / 2;double G = 0.5 * (1 - Math.sqrt(Math.pow(meanC, 7) / (Math.pow(meanC, 7) + Math.pow(25, 7))));double a1Prime = a1 * (1 + G);double a2Prime = a2 * (1 + G);double C1Prime = Math.sqrt(a1Prime * a1Prime + b1 * b1);double C2Prime = Math.sqrt(a2Prime * a2Prime + b2 * b2);double meanCPrime = (C1Prime + C2Prime) / 2;double h1Prime = Math.toDegrees(Math.atan2(b1, a1Prime));if (h1Prime < 0) h1Prime += 360;double h2Prime = Math.toDegrees(Math.atan2(b2, a2Prime));if (h2Prime < 0) h2Prime += 360;double deltaHPrime;if (Math.abs(h1Prime - h2Prime) <= 180) {deltaHPrime = h2Prime - h1Prime;} else if (h2Prime <= h1Prime) {deltaHPrime = h2Prime - h1Prime + 360;} else {deltaHPrime = h2Prime - h1Prime - 360;}double deltaH = 2 * Math.sqrt(C1Prime * C2Prime) * Math.sin(Math.toRadians(deltaHPrime / 2));double meanHPrime;if (Math.abs(h1Prime - h2Prime) <= 180) {meanHPrime = (h1Prime + h2Prime) / 2;} else if (h1Prime + h2Prime < 360) {meanHPrime = (h1Prime + h2Prime + 360) / 2;} else {meanHPrime = (h1Prime + h2Prime - 360) / 2;}double T = 1 - 0.17 * Math.cos(Math.toRadians(meanHPrime - 30))+ 0.24 * Math.cos(Math.toRadians(2 * meanHPrime))+ 0.32 * Math.cos(Math.toRadians(3 * meanHPrime + 6))- 0.20 * Math.cos(Math.toRadians(4 * meanHPrime - 63));double SL = 1 + (0.015 * Math.pow(meanL - 50, 2)) / Math.sqrt(20 + Math.pow(meanL - 50, 2));double SC = 1 + 0.045 * meanCPrime;double SH = 1 + 0.015 * meanCPrime * T;double RT = -2 * Math.sqrt(Math.pow(meanCPrime, 7) / (Math.pow(meanCPrime, 7) + Math.pow(25, 7)))* Math.sin(Math.toRadians(60 * Math.exp(-Math.pow((meanHPrime - 275) / 25, 2))));double deltaE = Math.sqrt(Math.pow(deltaL / SL, 2)+ Math.pow(deltaHPrime / SC, 2)+ Math.pow(deltaH / SH, 2)+ RT * (deltaCPrime / SC) * (deltaH / SH));return deltaE;
}

3.4.3 颜色相似度判断

public static boolean areColorsSimilar(Scalar color1, Scalar color2, double threshold, int colorSpace) {// 转换为指定颜色空间Mat mat1 = new Mat(1, 1, CvType.CV_8UC3, color1);Mat mat2 = new Mat(1, 1, CvType.CV_8UC3, color2);if (colorSpace != Imgproc.COLOR_BGR2RGB) {Imgproc.cvtColor(mat1, mat1, colorSpace);Imgproc.cvtColor(mat2, mat2, colorSpace);}Scalar c1 = new Scalar(mat1.get(0, 0));Scalar c2 = new Scalar(mat2.get(0, 0));// 根据颜色空间选择合适的比较方法if (colorSpace == Imgproc.COLOR_BGR2Lab) {double deltaE = colorDistanceCIEDE2000(c1, c2);return deltaE < threshold;} else {double distance = colorDistanceEuclidean(c1, c2);return distance < threshold;}
}

3.5 颜色对比应用实例

3.5.1 屏幕取色器实现

public class ColorPickerView extends View {private Bitmap mBitmap;private int mSelectedX = -1;private int mSelectedY = -1;private OnColorSelectedListener mListener;public interface OnColorSelectedListener {void onColorSelected(int color);}public ColorPickerView(Context context) {super(context);}public ColorPickerView(Context context, AttributeSet attrs) {super(context, attrs);}public void setBitmap(Bitmap bitmap) {mBitmap = bitmap;invalidate();}public void setOnColorSelectedListener(OnColorSelectedListener listener) {mListener = listener;}@Overrideprotected void onDraw(Canvas canvas) {super.onDraw(canvas);if (mBitmap != null) {// 绘制图像Rect src = new Rect(0, 0, mBitmap.getWidth(), mBitmap.getHeight());Rect dst = new Rect(0, 0, getWidth(), getHeight());canvas.drawBitmap(mBitmap, src, dst, null);// 绘制选择点标记if (mSelectedX >= 0 && mSelectedY >= 0) {Paint paint = new Paint();paint.setColor(Color.WHITE);paint.setStyle(Paint.Style.STROKE);paint.setStrokeWidth(3);canvas.drawCircle(mSelectedX, mSelectedY, 20, paint);// 获取并显示颜色值int color = getPixelColor(mBitmap, mSelectedX, mSelectedY);paint.setStyle(Paint.Style.FILL);paint.setColor(color);canvas.drawRect(getWidth() - 100, 20, getWidth() - 20, 100, paint);paint.setColor(Color.BLACK);paint.setTextSize(30);canvas.drawText(String.format("#%06X", (0xFFFFFF & color)), getWidth() - 180, 70, paint);}}}@Overridepublic boolean onTouchEvent(MotionEvent event) {if (mBitmap == null) return super.onTouchEvent(event);int x = (int)(event.getX() * mBitmap.getWidth() / getWidth());int y = (int)(event.getY() * mBitmap.getHeight() / getHeight());if (x >= 0 && x < mBitmap.getWidth() && y >= 0 && y < mBitmap.getHeight()) {mSelectedX = x;mSelectedY = y;if (mListener != null) {int color = getPixelColor(mBitmap, x, y);mListener.onColorSelected(color);}invalidate();return true;}return super.onTouchEvent(event);}private int getPixelColor(Bitmap bitmap, int x, int y) {if (x < 0 || y < 0 || x >= bitmap.getWidth() || y >= bitmap.getHeight()) {return Color.TRANSPARENT;}return bitmap.getPixel(x, y);}
}

3.5.2 颜色匹配检测

public class ColorMatchDetector {private Scalar mTargetColor;private double mThreshold;private int mColorSpace;public ColorMatchDetector(Scalar targetColor, double threshold, int colorSpace) {mTargetColor = targetColor;mThreshold = threshold;mColorSpace = colorSpace;}public Mat findColorRegions(Mat inputImage) {// 转换为目标颜色空间Mat converted = new Mat();Imgproc.cvtColor(inputImage, converted, mColorSpace);// 创建目标颜色矩阵Mat targetMat = new Mat(inputImage.size(), converted.type(), mTargetColor);// 计算差异Mat diff = new Mat();Core.absdiff(converted, targetMat, diff);// 分割通道List<Mat> channels = new ArrayList<>();Core.split(diff, channels);// 计算总差异Mat totalDiff = new Mat(channels.get(0).size(), CvType.CV_32F);for (Mat channel : channels) {Mat floatChannel = new Mat();channel.convertTo(floatChannel, CvType.CV_32F);Core.add(totalDiff, floatChannel.mul(floatChannel), totalDiff);}Core.sqrt(totalDiff, totalDiff);// 创建掩码Mat mask = new Mat();Core.inRange(totalDiff, new Scalar(0), new Scalar(mThreshold), mask);// 清理小区域Mat kernel = Imgproc.getStructuringElement(Imgproc.MORPH_ELLIPSE, new Size(5, 5));Imgproc.morphologyEx(mask, mask, Imgproc.MORPH_OPEN, kernel);Imgproc.morphologyEx(mask, mask, Imgproc.MORPH_CLOSE, kernel);return mask;}public List<Rect> findColorContours(Mat inputImage) {Mat mask = findColorRegions(inputImage);// 查找轮廓List<MatOfPoint> contours = new ArrayList<>();Mat hierarchy = new Mat();Imgproc.findContours(mask, contours, hierarchy, Imgproc.RETR_EXTERNAL, Imgproc.CHAIN_APPROX_SIMPLE);// 转换为矩形List<Rect> rects = new ArrayList<>();for (MatOfPoint contour : contours) {Rect rect = Imgproc.boundingRect(contour);if (rect.area() > 100) { // 忽略小区域rects.add(rect);}}return rects;}
}

四、指定区域OCR内容提取

4.1 OCR技术简介

OCR（Optical Character Recognition，光学字符识别）是将图像中的文字转换为可编辑文本的技术。在Android中实现OCR通常有以下几种方式：

1. 使用Tesseract OCR引擎
2. 使用Google ML Kit文本识别API
3. 使用第三方OCR服务API

4.2 Tesseract OCR集成

4.2.1 添加Tesseract依赖

在build.gradle中添加：

dependencies {implementation 'com.rmtheis:tess-two:9.1.0'
}

4.2.2 初始化Tesseract

public class TessOCR {private TessBaseAPI mTess;public TessOCR(Context context, String language) {mTess = new TessBaseAPI();// 训练数据路径String datapath = context.getFilesDir() + "/tesseract/";File dir = new File(datapath + "tessdata/");if (!dir.exists()) {dir.mkdirs();}// 检查训练数据文件是否存在File tessdataFile = new File(datapath + "tessdata/" + language + ".traineddata");if (!tessdataFile.exists()) {try {// 从assets复制训练数据InputStream in = context.getAssets().open("tessdata/" + language + ".traineddata");OutputStream out = new FileOutputStream(tessdataFile);byte[] buffer = new byte[1024];int read;while ((read = in.read(buffer)) != -1) {out.write(buffer, 0, read);}in.close();out.flush();out.close();} catch (IOException e) {e.printStackTrace();}}mTess.init(datapath, language);}public String recognizeText(Bitmap bitmap) {if (bitmap == null) return "";mTess.setImage(bitmap);return mTess.getUTF8Text();}public String recognizeText(Mat mat) {if (mat.empty()) return "";Bitmap bitmap = Bitmap.createBitmap(mat.cols(), mat.rows(), Bitmap.Config.ARGB_8888);Utils.matToBitmap(mat, bitmap);return recognizeText(bitmap);}public void destroy() {if (mTess != null) {mTess.end();}}
}

4.3 图像预处理优化OCR结果

4.3.1 基本预处理流程

public static Mat prepareImageForOCR(Mat src) {if (src.empty()) return new Mat();Mat processed = new Mat();// 1. 转换为灰度图像Imgproc.cvtColor(src, processed, Imgproc.COLOR_BGR2GRAY);// 2. 应用自适应阈值Imgproc.adaptiveThreshold(processed, processed, 255, Imgproc.ADAPTIVE_THRESH_GAUSSIAN_C,Imgproc.THRESH_BINARY, 11, 2);// 3. 降噪Imgproc.medianBlur(processed, processed, 3);// 4. 锐化Mat kernel = new Mat(3, 3, CvType.CV_32F) {{put(0, 0, 0);put(0, 1, -1);put(0, 2, 0);put(1, 0, -1);put(1, 1, 5);put(1, 2, -1);put(2, 0, 0);put(2, 1, -1);put(2, 2, 0);}};Imgproc.filter2D(processed, processed, -1, kernel);return processed;
}

4.3.2 高级预处理技术

public static Mat advancedOCRPreprocessing(Mat src) {if (src.empty()) return new Mat();Mat processed = new Mat();// 1. 转换为灰度图像Imgproc.cvtColor(src, processed, Imgproc.COLOR_BGR2GRAY);// 2. 应用CLAHE (对比度受限的自适应直方图均衡化)CLAHE clahe = Imgproc.createCLAHE();clahe.setClipLimit(2);clahe.apply(processed, processed);// 3. 非局部均值去噪Photo.fastNlMeansDenoising(processed, processed, 10, 7, 21);// 4. 边缘增强Mat edges = new Mat();Imgproc.Sobel(processed, edges, CvType.CV_8U, 1, 1);Core.add(processed, edges, processed);// 5. 局部自适应阈值Imgproc.adaptiveThreshold(processed, processed, 255, Imgproc.ADAPTIVE_THRESH_GAUSSIAN_C,Imgproc.THRESH_BINARY, 15, 5);// 6. 形态学操作去除小噪点Mat kernel = Imgproc.getStructuringElement(Imgproc.MORPH_RECT, new Size(2, 2));Imgproc.morphologyEx(processed, processed, Imgproc.MORPH_OPEN, kernel);return processed;
}

4.4 指定区域OCR实现

4.4.1 从指定区域提取文本

public class RegionOCR {private TessOCR mTessOCR;public RegionOCR(Context context, String language) {mTessOCR = new TessOCR(context, language);}public String extractTextFromRegion(Bitmap srcBitmap, Rect region) {if (srcBitmap == null || region == null) return "";// 裁剪指定区域Bitmap cropped = Bitmap.createBitmap(srcBitmap, region.left, region.top,region.width(), region.height());// 预处理图像Mat mat = new Mat();Utils.bitmapToMat(cropped, mat);mat = prepareImageForOCR(mat);// 执行OCRreturn mTessOCR.recognizeText(mat);}public String extractTextFromRegion(Mat srcMat, Rect region) {if (srcMat.empty() || region == null) return "";// 调整区域确保在图像范围内Rect adjusted = new Rect(Math.max(0, region.x),Math.max(0, region.y),Math.min(srcMat.cols() - region.x, region.width),Math.min(srcMat.rows() - region.y, region.height));// 裁剪指定区域Mat cropped = new Mat(srcMat, adjusted);// 预处理图像Mat processed = prepareImageForOCR(cropped);// 执行OCRreturn mTessOCR.recognizeText(processed);}public void destroy() {if (mTessOCR != null) {mTessOCR.destroy();}}
}