Multiple Adaptive Derivative Passive Image Processing Approach to Unsharp Masking Restoration for CNN on Low Light Images

Most researchers thought face detection challenges are distinguishing faces against the background images and coping with multiple face poses whereby every face will only have an eye, half of a nose and half of a mouth when a face is taken from one side profile image. The actual challenge of face detection is when there is no light or luminant to shine on the face when the image is taken. This challenge is also known as low light illumination. These days, CNNs have been used as the basis of deep learning frameworks in various research works. This paper compares our restoration process against existing non-CNN-based and CNN-based methods. It evaluates whether Retinex and Bilateral theory still can be applied in enhancing and solving face detection in low light. When an image goes through multiple adaptive changes, multiple times as input and output, it is no longer appropriate to call it an output but rather a derivative. Thus, our proposed output is labelled as Multiple Adaptive Derivative and this paper evaluates the performance of the proposed Multiple Adaptive Derivative Passive Image Processing (MADPIP) restoration and enhancement processes for images taken in low light using Logarithmic, CLAHE and Unsharp Masking. Apart from contrast and edge quality, MADPIP’s integrity, image entropy, visual edge and time processing are also measured.