OpenCV in advanced Android development: light source detection

In the last Android development tutorial we’ve covered drawing a histogram using OpenCV. This time we’ll show you how to detect light sources in an image. This feature comes in very handy when building an Android app that has to perform quality checks on user-generated photographs (e.g. sending documents for proof).

Our goal is to find the light source point and draw there a circle in an Android app. The easiest way to get it done is to detect the localization where the greatest white pixel value is. We can do it by using the Core.minMaxLoc method, which takes only one parameter – the array.

In order to detect light source with OpenCV in our Android application we need to perform 4 steps:

1. Read the bitmap
2. Transform it to grayscale
3. Apply Gaussian blur
4. Calculate maxLoc by using the minMaxLoc method

    private void detectLight(Bitmap bitmap, double gaussianBlurValue) {
        Mat rgba = new Mat();
        Utils.bitmapToMat(bitmap, rgba);

        Mat grayScaleGaussianBlur = new Mat();
        Imgproc.cvtColor(rgba, grayScaleGaussianBlur, Imgproc.COLOR_BGR2GRAY);
        Imgproc.GaussianBlur(grayScaleGaussianBlur, grayScaleGaussianBlur, new Size(gaussianBlurValue, gaussianBlurValue), 0);

        Core.MinMaxLocResult minMaxLocResultBlur = Core.minMaxLoc(grayScaleGaussianBlur);
        Imgproc.circle(rgba, minMaxLocResultBlur.maxLoc, 30, new Scalar(255), 3);

        // Don't do that at home or work it's for visualization purpose.
        Bitmap resultBitmap = Bitmap.createBitmap(rgba.cols(), rgba.rows(), Bitmap.Config.ARGB_8888);
        Utils.matToBitmap(rgba, resultBitmap);
        BitmapHelper.showBitmap(this, resultBitmap, detectLightImageView);

        Bitmap blurryBitmap = Bitmap.createBitmap(grayScaleGaussianBlur.cols(), grayScaleGaussianBlur.rows(), Bitmap.Config.ARGB_8888);
        Utils.matToBitmap(grayScaleGaussianBlur, blurryBitmap);
        BitmapHelper.showBitmap(this, blurryBitmap, gaussianBlurImageView);

    }

Extract data and apply transformations

// Covert our bitmap into an array
Mat rgba = new Mat();
Utils.bitmapToMat(bitmap, rgba);

// Convert our rgba array to greyscale by using Imgproc.cvtColor ()
// Apply Gaussian blur (for the last one we recommend Imgproc.GaussianBlur()).
Mat grayScaleGaussianBlur = new Mat();
Imgproc.cvtColor(rgba, grayScaleGaussianBlur, Imgproc.COLOR_BGR2GRAY);
Imgproc.GaussianBlur(grayScaleGaussianBlur, grayScaleGaussianBlur, new Size(gaussianBlurValue, gaussianBlurValue), 0);

Imgproc.cvtColor parameters:

• Mat scr – the source array (in our case rgba)
• Mat dst – the array in which we will save the result of the conversion (in our case grayScaleGaussianBlur)
• int code – code identifying the type of conversion

Imgproc.GaussianBlur parameters:

• Mat scr – source array
• Mat dst– output array (the blurring result will be saved to it)
• Size ksize – Gaussian kernel size. ksize.width and ksize.height can differ but they both must be positive and odd. Or, they can be zero’s and then they are computed from sigma
• double sigmaX – Gaussian kernel standard deviation in X direction

Analyze the image and display the output

By using the Core.minLoc() method we can calculate the brightest spot in our photo and draw a circle around it using Imgproc.circle().

Core.MinMaxLocResult minMaxLocResultBlur = Core.minMaxLoc(grayScaleGaussianBlur);
Imgproc.circle(rgba, minMaxLocResultBlur.maxLoc, 30, new Scalar(255), 3);

Finally, we display two pictures, one with the circle and the other one with the transformations applied (greyscale + blur).

// Don't do that at home or work it's for visualization purpose.
Bitmap resultBitmap = Bitmap.createBitmap(rgba.cols(), rgba.rows(), Bitmap.Config.ARGB_8888);
Utils.matToBitmap(rgba, resultBitmap);
BitmapHelper.showBitmap(this, resultBitmap, detectLightImageView);

Bitmap blurryBitmap = Bitmap.createBitmap(grayScaleGaussianBlur.cols(), grayScaleGaussianBlur.rows(), Bitmap.Config.ARGB_8888);
Utils.matToBitmap(grayScaleGaussianBlur, blurryBitmap);
BitmapHelper.showBitmap(this, blurryBitmap, gaussianBlurImageView);

This method may not be perfect, but the effects are really good and the built Android application can detect the light now.

OpenCV: light source detection

A code sample that we’ve used in a commercial Android application can be found on our GitHub page.