从一组图片中检测（并去除）划痕

Question

To simplify, I have a set of pictures shot from a camera. This camera was behind an acrylic dome and this dome has scratches on it. The scratches are very apparent in the pictures. With opencv, I'm trying to detect and remove these scratches. My first approach is to multiply a subset of the pictures to create a mask, than apply the inpaint mask. Am I on the right track? Is there a treatment I can do on the images before multiplying them?

EDIT: Here is an image highlighting the scratches

Answer 1

After a few tests, I came with a reliable way to eliminate recurring scratches and dust spots on a set of images. First, I take a subset of 10 to 15 pictures, open them, and apply sobel derivatives to extract the contours.

        private Image<Gray, byte> GetImageContours(String strImage)
        {
            Image<Gray, Byte> img = new Image<Gray, byte>(strImage);
            Image<Gray, float> gradx = img.Sobel(1, 0, 5);
            Image<Gray, Byte> gradxconv = gradx.ConvertScale<Byte>(1, 0);
            Image<Gray, float> grady = img.Sobel(0, 1, 5);
            Image<Gray, Byte> gradyconv = grady.ConvertScale<Byte>(1, 0);
            Image<Gray, Byte> imgContours = gradxconv.AddWeighted(gradyconv, 0.5, 0.5, 0);

            img.Dispose();
            gradx.Dispose();
            grady.Dispose();
            gradxconv.Dispose();
            gradyconv.Dispose();

            return imgContours;
        }

Then, I convert the images back to a 16bit depth, then multiply all of them. This should give me an image of the scratches only. To create a mask of the scratches, I erode this image one time, then dilate it 10 times. After that, I convert it to black and white by running a TresholdBinary with a treshold value of 40.

The final step is to use the Inpaint function on all the set of images with the scratch mask.

            Image<Gray, float> _img1 = img1.Convert<Gray, float>();
            Image<Gray, float> _img2 = img2.Convert<Gray, float>();
            Image<Gray, float> _img3 = img3.Convert<Gray, float>();
            Image<Gray, float> _img4 = img4.Convert<Gray, float>();
            Image<Gray, float> _img5 = img5.Convert<Gray, float>();
            Image<Gray, float> _img6 = img6.Convert<Gray, float>();
            Image<Gray, float> _img7 = img7.Convert<Gray, float>();
            Image<Gray, float> _img8 = img8.Convert<Gray, float>();
            Image<Gray, float> _img9 = img9.Convert<Gray, float>();
            Image<Gray, float> _img10 = img10.Convert<Gray, float>();
            Image<Gray, float> _img11 = img11.Convert<Gray, float>();
            Image<Gray, float> _img12 = img12.Convert<Gray, float>();
            Image<Gray, float> imgMask = _img1.Mul(_img2).Mul(_img3).Mul(_img4).Mul(_img5).Mul(_img6).Mul(_img7).Mul(_img8).Mul(_img9).Mul(_img10).Mul(_img11).Mul(_img12).Erode(1);
            Image<Gray, Byte> imgMask2 = imgMask.Convert<Gray, Byte>();
            Image<Gray, Byte> imgMask3 = imgMask2.Dilate(10).ThresholdBinary(new Gray(40), new Gray(255));

            //img is one of the original images to remove scratches on
            ImageViewer viewer3 = new ImageViewer(img.InPaint(imgMask3, 3), "image3");
            viewer3.Show();

Here are sample images taken at each step:

original image

contours of this image

all the contours multiplied

the mask eroded and dilated

the final result