How to normalize image in openCV?

Question

I am totally new in openCV and stuck at one point.

I have a grey scale image but I need to normalize this image. For example: I have a cv::mat image which may have a matrix containing some values and each index. As I have gray image it may contain only 1 value per index. Now I need to divide each value by 255.

Is there any method or facility available in openCV in C++? For this scenario, I believe the method I want to use is called normalization in openCV?

cv::Mat originalMat = [OSInference cvMatFromUIImage:imgBeforeProccessing];
cv::Mat img2;

cv::cvtColor(originalMat, img2,CV_BGR2GRAY);
cv::resize(img2, img2, cv::Size(128, 128), 0, 0, CV_INTER_CUBIC);
cv::Mat img3;

Now how to normalise (this means to divide each value in the matrix by 255)??

I am converting the mat image to an iOS image as follows:

+ (UIImage *)UIImageFromCVMat:(cv::Mat)cvMat{
    NSData *data = [NSData dataWithBytes:cvMat.data length:cvMat.elemSize()*cvMat.total()];

    CGColorSpaceRef colorspace;

    if (cvMat.elemSize() == 1) {
        colorspace = CGColorSpaceCreateDeviceGray();
    }else{
        colorspace = CGColorSpaceCreateDeviceRGB();
    }

    CGDataProviderRef provider = CGDataProviderCreateWithCFData((__bridge CFDataRef)data);

    // Create CGImage from cv::Mat
    CGImageRef imageRef = CGImageCreate(cvMat.cols, cvMat.rows, 8, 8 * cvMat.elemSize(), cvMat.step[0], colorspace, kCGImageAlphaNone | kCGBitmapByteOrderDefault, provider, NULL, false, kCGRenderingIntentDefault);

    // get uiimage from cgimage
    UIImage *finalImage = [UIImage imageWithCGImage:imageRef];
    CGImageRelease(imageRef);
    CGDataProviderRelease(provider);
    CGColorSpaceRelease(colorspace);

    return finalImage;
}

Solution 1

OpenCV has a function does exactly what you want. It's called convertTo.

cv::Mat img3;
img2.convertTo(img3, CV_32F, 1.0 / 255, 0);

Solution 2

To normalize cv::Mat you can use cv::normalize. I write some code to help you.

uchar data[] = {0, 63, 127, 255};
cv::Mat im(2, 2, CV_8UC1, data), output;
# this what you need 0 -> min value after norm
# 1 -> max value after nom
# cv::NORM_MINMAX normalize for min and max values
cv::normalize(im, output, 0, 1, cv::NORM_MINMAX);
std::cout 
    << im     << '\n'
    << output << '\n';
#==================================================
# output
[  0,  63;
   127, 255]
[  0,   0;
   0,   1]

Comments

[New iOS Dev]

is normalisation concept is same what i want?

[New iOS Dev]

if I use my grey image converting to somewhat colour image (yellowish) is this possible?

[slawekwin]

@sss I don't quite understand what you mean

[New iOS Dev]

As i am converting colour image to grey first then applying convertTo method....but after this my image turns yellowish ...as per my consideration grey image should remain grey even after convertTo method

[slawekwin]

@sss how do you display the converted image? resulting values are in range <0, 1> and cv::imshow should display it the same as before conversion

[New iOS Dev]

Please find my updated question..i am just calling method to convert image to IOS UIImage . like this img2.convertTo(img3, CV_32F, 1.0 / 255, 0); processedImageForModel = [OSInference UIImageFromCVMat:img3];

[New iOS Dev]

UIImageFromCVMat . is my custom method which I write in question

[Miki]

The answer is correct. I think that the issue is that UIImageFromCVMat handles only CV_8U images, while this is CV_32F. So, for visualization purposes, the image should be of type CV_8U with values in [0,255]

[slawekwin]

@sss I believe it would behave the same as the solution in my answer, but it also has to calculate the factor (1/255) to multiply all values by itself (see docs)

[New iOS Dev]

should I Convert CV_8U to CV_32F in my custom method? or is there openCV method to do this?

[Miki]

@slawekwin This is not the same as your question (and is in fact not correct). It's the same only if the input image happens to have both 0 and 255 values, otherwise the scaling factor is not 1/255

[Miki]

The easiest way is to pass to UIImageFromCVMat matrices of the correct type. So you can do: Mat matToShow; img3.convertTo(matToShow, CV_8U, 255); However, this is the opposite of the operation you just did. So you can visualize directly img2, and keep img3 for further processing

[slawekwin]

@sss I see you use CGImageCreate function for your conversion. Read the doc for it, I think specyfing CGColorSpaceCreateDeviceGray color space and 32 for bitsPerComponent and bitsPerPixel might help

[New iOS Dev]

@Miki is this correct what I am trying to do based on your information . //------ Open CV preproccessing cv::Mat originalMat = [OSInference cvMatFromUIImage:imgBeforeProccessing]; cv::Mat img2; cv::cvtColor(originalMat, img2,CV_BGR2GRAY); cv::resize(img2, img2, cv::Size(128, 128), 0, 0, CV_INTER_CUBIC); cv::Mat img3; img2.convertTo(img3, CV_32F, 1.0 / 255.0, 0); img3.convertTo(img3, CV_8U, 255); processedImageForModel = [OSInference UIImageFromCVMat:img3];

[Miki]

Yes, but img3 will be exactly equal to img2. I think you can do: cv::Mat img2; cv::cvtColor(originalMat, img2,CV_BGR2GRAY); cv::resize(img2, img2, cv::Size(128, 128), 0, 0, CV_INTER_CUBIC); cv::Mat img3; img2.convertTo(img3, CV_32F, 1.0 / 255.0, 0); processedImageForModel = [OSInference UIImageFromCVMat:img2]; and keep img3 for further processing.

[Miki]

However isn't clear why you need to scale by 1/255 at all ;)

[New iOS Dev]

I don't have further processing just needed iOS image from that cvmat which i am passing to UIImageFromCVMat method

[Miki]

Then just cv::Mat img2; cv::cvtColor(originalMat, img2,CV_BGR2GRAY); processedImageForModel = [OSInference UIImageFromCVMat:img2]