Harr is a feature detection technique which is used to highlight the important features in an image which will contribute to final CNN model during training of the models. It is like a kernel or filter detector consisting of black and white pixels which is convolved with the incoming image coverted into gray scale image. Each feature is a single value which is resulted by subtracting the sum of pixels under white rectangle from the sum of pixels under black rectangle.

In the below snap we have 4 different types of features which will be convolved with the image. Based on the highest value of feature in the output image after applying calculation we can identify whether the feature is important in order to identify the face and non-face.

#machine-learning #viola-jones #data analysis