使用OpenCV cascadeClassifier.detectMultiScale处理大图时内存不足问题求助

问题描述 #

我用Java编写了以下检测图片是否含有人脸的代码：

public boolean isContainsFace(String path){
        Mat matImage = Imgcodecs.imread(path);

        MatOfRect faceDetections = new MatOfRect();
        cascadeClassifier.detectMultiScale(matImage, faceDetections);

        return !faceDetections.empty();
    }

处理大尺寸图片时，触发了内存不足错误：

E/cv::error(): OpenCV(4.6.0-dev) Error: Insufficient memory (Failed to allocate 1281229312 bytes) in OutOfMemoryError, file E:/OpenCV/opencv/modules/core/src/alloc.cpp, line 73
E/org.opencv.objdetect: objdetect::detectMultiScale_15() caught cv::Exception: OpenCV(4.6.0-dev) E:/OpenCV/opencv/modules/core/src/alloc.cpp:73: error: (-4:Insufficient memory) Failed to allocate 1281229312 bytes in function 'OutOfMemoryError'
E/AndroidRuntime: FATAL EXCEPTION: main
    Process: com.example.findyourselfinthephoto, PID: 25403
    CvException [org.opencv.core.CvException: cv::Exception: OpenCV(4.6.0-dev) E:/OpenCV/opencv/modules/core/src/alloc.cpp:73: error: (-4:Insufficient memory) Failed to allocate 1281229312 bytes in function 'OutOfMemoryError'
    ]
        at org.opencv.objdetect.CascadeClassifier.detectMultiScale_5(Native Method)
        at org.opencv.objdetect.CascadeClassifier.detectMultiScale(CascadeClassifier.java:195)
        at com.example.findyourselfinthephoto.Helpers.PhotoHelper.isContainsFace(PhotoHelper.java:41)
        at com.example.findyourselfinthephoto.Fragments.Home.onRequestHandlePathOz(Home.java:164)
        at br.com.onimur.handlepathoz.utils.HandlePathOzUtils$getRealPath$1.invokeSuspend(HandlePathOzUtils.kt:107)
        at kotlin.coroutines.jvm.internal.BaseContinuationImpl.resumeWith(ContinuationImpl.kt:33)
        at kotlinx.coroutines.internal.ScopeCoroutine.afterResume(Scopes.kt:33)
        at kotlinx.coroutines.AbstractCoroutine.resumeWith(AbstractCoroutine.kt:102)
        at kotlin.coroutines.jvm.internal.BaseContinuationImpl.resumeWith(ContinuationImpl.kt:46)
        at kotlinx.coroutines.DispatchedTask.run(DispatchedTask.kt:106)
        at android.os.Handler.handleCallback(Handler.java:938)
        at android.os.Handler.dispatchMessage(Handler.java:99)
        at android.os.Looper.loop(Looper.java:223)
        at android.app.ActivityThread.main(ActivityThread.java:7656)
        at java.lang.reflect.Method.invoke(Native Method)
        at com.android.internal.os.RuntimeInit$MethodAndArgsCaller.run(RuntimeInit.java:592)
        at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:947)
        Suppressed: kotlinx.coroutines.DiagnosticCoroutineContextException: [StandaloneCoroutine{Cancelling}@e6b34af, Dispatchers.Main]

解决方案 #

1. 缩放图片后再处理 #

大尺寸图片加载到内存中会占用大量空间，尤其是OpenCV的Mat对象直接映射像素数据。可以先将图片缩放到合适尺寸（比如长边限制在1000像素以内），再执行检测：

public boolean isContainsFace(String path){
        Mat matImage = Imgcodecs.imread(path);
        
        // 计算缩放比例，限制长边最大尺寸
        int maxDim = 1000;
        int height = matImage.rows();
        int width = matImage.cols();
        double scale = 1.0;
        if (width > maxDim || height > maxDim) {
            scale = (double) maxDim / Math.max(width, height);
            Imgproc.resize(matImage, matImage, new Size(width * scale, height * scale));
        }

        MatOfRect faceDetections = new MatOfRect();
        cascadeClassifier.detectMultiScale(matImage, faceDetections);
        
        // 手动释放内存
        matImage.release();
        faceDetections.release();

        return !faceDetections.empty();
    }

2. 及时释放Mat资源 #

OpenCV的Mat对象在Android环境下依赖原生内存，若不手动调用release()，容易引发内存泄漏。处理完成后必须主动释放相关资源。

3. 使用灰度图检测 #

人脸检测不需要彩色信息，将图片转为灰度图可减少2/3的内存占用（彩色图为3通道，灰度图为1通道）：

public boolean isContainsFace(String path){
        Mat matImage = Imgcodecs.imread(path);
        
        // 转为灰度图并释放原彩色图内存
        Mat grayMat = new Mat();
        Imgproc.cvtColor(matImage, grayMat, Imgproc.COLOR_BGR2GRAY);
        matImage.release();

        // 结合缩放处理
        int maxDim = 1000;
        int height = grayMat.rows();
        int width = grayMat.cols();
        double scale = 1.0;
        if (width > maxDim || height > maxDim) {
            scale = (double) maxDim / Math.max(width, height);
            Imgproc.resize(grayMat, grayMat, new Size(width * scale, height * scale));
        }

        MatOfRect faceDetections = new MatOfRect();
        cascadeClassifier.detectMultiScale(grayMat, faceDetections);
        
        grayMat.release();
        faceDetections.release();

        return !faceDetections.empty();
    }

4. 优化detectMultiScale参数 #

调整检测参数，增大scaleFactor、minNeighbors，减少不必要的计算，同时降低内存消耗：

// 示例参数，可根据实际场景调整
cascadeClassifier.detectMultiScale(grayMat, faceDetections, 1.1, 3, 0, new Size(30, 30));

内容的提问来源于stack exchange，提问作者Tamada