The main aim of this assignment is to perform face detection by using the integral image. This is to be done by detecting eye area in faces through convolving a kernel that is designed to detect that area. The assignment is structured as follows: 1. Calculate the integral image and calculate the local sum 2. Detect the eye area. Detect Eye Area In this part, you are asked to implement two functions as follows: 1. DetectEye: • Input: the integral image and the kernel width. • Output: Coordinates (i, j) which represent the position of the maximum score achieved after convoluting the kernel. • Description: This method is responsible for calculating the local sums for the kernel provided in order to detect the eyes in the face image, where it gives the maximum score in the area where the eyes are present through convolving that kernel. To convolve the kernel, the width of the kernel is provided as an input parameter and the height of the kernel is calculate from the width by the equation: m = 0.15*n, where m defines the height of the kernel and n defines the kernel width. Next, calculate the maximum score while convolving the kernel by calculating the different local sums of the different areas in the kernel and then save the position of the maximum score. ➢ The kernel is shown below including the points representing each area (P1 … P14) in order to calculate seven different local sums (LS1 … LS7). The kernel is designed to match the area of the face showing the eyes, the eye brows and the forehead. ➢ The gray color refers to a neglection area (zeros), the white color refers to ones in the kernel and the black area refers to negative ones in the kernel.2. ExtractDetectedEye: • Input: Takes the image itself, the maximum position retrieved from DetectEye method and the kernel width size • Output: a 2D array representing the image drawn. • Description: This method is responsible to extract the eye area itself as detected from the first method. As we have the position of the maximum score, then draw around that position the same kernel size that detects the eye from the whole image. So, it is only retrieving the pixels of interest which are the eyes in the same range of the kernel size in terms of width and height.
