图像处理软件开发记录(五) 图像特效（素描、油画）

左手的ㄟ右手 2022-08-26 08:37 261阅读 0赞

专栏地址：[http://blog.csdn.net/column/details/imagep.html][http_blog.csdn.net_column_details_imagep.html]

本篇文章主要记录一下图像处理软件中的图像特效（素描和油画）的实现过程。

**图像素描效果**

图像素描的实现原理其实很简单，主要是利用边缘检测滤波器来实现。

可供选择的边缘检测滤波器有很多，常用的有Sobel、Scharr、Laplacian以及Canny滤波。本文主要是利用**[Laplacian][]**来实现素描化，它的效果相对于Sobel和Scharr更加相像素描效果。

Laplacian算子是一个二阶导数算子，具有各向同性，即与坐标轴方向无关，坐标轴旋转后梯度结果不变。但是，它对噪声比较敏感，所以我们这里先利用[中值滤波器][Link 1]对图像进行去噪处理。

Laplacian常用的卷积核来近似，如下图所示：

![SouthEast][]

**Code:**

void MainWindow::on_actionSketch_triggered()
    {
        cv::Mat gray;
        cv::cvtColor(image,gray,CV_BGR2GRAY);
        const int MEDIAN_BLUR_FILTER_SIZE = 7;
        cv::medianBlur(gray,gray,MEDIAN_BLUR_FILTER_SIZE);
    
        cv::Mat edges;
        const int LAPLACIAN_FILTER_SIZE = 5;
        cv::Laplacian(gray,edges,CV_8U,LAPLACIAN_FILTER_SIZE);
    
        cv::Mat mask;
        const int EDGE_THRESHOLD = 50;
        cv::threshold(edges,mask,EDGE_THRESHOLD,255,cv::THRESH_BINARY_INV);
        QImage simg = QImage((const unsigned char*)(mask.data),mask.cols,mask.rows,
                             QImage::Format_Indexed8);
        //ui->Imagedisplaylabel->setPixmap(QPixmap::fromImage(simg));
        //ui->Imagedisplaylabel->setScaledContents(true);
        //ui->Imagedisplaylabel->resize(ui->Imagedisplaylabel->width(),ui->Imagedisplaylabel->height());
    
    }

效果：（华大一景）

![SouthEast 1][]

![SouthEast 2][]

**图像油画效果**

关于图像的油画效果，本文是先利用双边滤波器（见[我的博客][Link 2]），将图像的区域推平，然后再将上面的素描效果叠加到经滤波后的图像上，如此即可实现图像的油画效果。

**Code：**

void MainWindow::on_actionCartoon_triggered()
    {
    
        cv::Mat gray;
        cv::cvtColor(image,gray,CV_BGR2GRAY);
        const int MEDIAN_BLUR_FILTER_SIZE = 7;
        cv::medianBlur(gray,gray,MEDIAN_BLUR_FILTER_SIZE);
    
        cv::Mat edges;
        const int LAPLACIAN_FILTER_SIZE = 5;
        cv::Laplacian(gray,edges,CV_8U,LAPLACIAN_FILTER_SIZE);
    
        cv::Mat mask ;
        const int EDGE_THRESHOLD = 90;
        cv::threshold(edges,mask,EDGE_THRESHOLD,255,cv::THRESH_BINARY_INV);
    
        cv::Size size = image.size();
        cv::Size smallSize;
        smallSize.width = size.width/2;
        smallSize.height = size.height/2;
        cv::Mat smallImg = cv::Mat(smallSize,CV_8UC3);
        cv::resize(image,smallImg,smallSize,0,0,cv::INTER_LINEAR);
    
        cv::Mat tmp = cv::Mat(smallSize,CV_8UC3);
        int repetitions = 9;
        for (int i = 0; i < repetitions; i++)
        {
            int ksize = 9;
            double sigmaColor = 9;
            double sigmaSpace = 7;
            cv::bilateralFilter(smallImg,tmp,ksize,sigmaColor,sigmaSpace);
            cv::bilateralFilter(tmp,smallImg,ksize,sigmaColor,sigmaSpace);
        }
    
        cv::Mat bigImg;
        cv::resize(smallImg,bigImg,image.size(),0,0,cv::INTER_LINEAR);
    
        cv::Mat cartoon= cv::Mat(image.size(),CV_8UC3);
        memset((char*)cartoon.data,0,cartoon.step * cartoon.rows);
    
        bigImg.copyTo(cartoon,mask);
    
        cv::imwrite("./mask1.jpg",cartoon);
        ui->textBrowser->clear();
        ui->textBrowser->append("<img src= ./mask1.jpg>");
    }

效果：（华大主楼）

![SouthEast 3][]

![SouthEast 4][]

**更多关于图像处理、机器视觉资料，请关注博客：[LinJM-机器视觉][LinJM-] 微博：[林建民-机器视觉][-]**

[http_blog.csdn.net_column_details_imagep.html]: http://blog.csdn.net/column/details/imagep.html
[Laplacian]: http://baike.baidu.com/view/3703285.htm
[Link 1]: http://zh.wikipedia.org/wiki/%E4%B8%AD%E5%80%BC%E6%BB%A4%E6%B3%A2%E5%99%A8
[SouthEast]: /images/20220824/1d733a41537d4c09b7fda65d1dd4ad5a.png
[SouthEast 1]: /images/20220824/af7e5ae703264f77b84152f52eae44dd.png
[SouthEast 2]: /images/20220824/a90de423cf624ffb87e70c9d70e82622.png
[Link 2]: http://blog.csdn.net/linj_m/article/details/21892471
[SouthEast 3]: /images/20220824/55e613c0ddc24cf3909b270c0f4ac6ec.png
[SouthEast 4]: /images/20220824/08ca1cbb7d464d9b9b64c05481638ba9.png
[LinJM-]: http://blog.csdn.net/linj_m
[-]: http://weibo.com/cengineer