Convenient sequential deep learning model by using CNN-QuadGRN for video noise removal and edge detection: A different perspective

Digital forensics is an important has a unique focus in forensic study. It deals with digital evidence captured by Webcam, Mobile camera, surveillance cameras and etc. Forensic outlines methodical and technical examinations of video facts, human testimonies, atmosphere, diurnal, persons involved and objects used. Resolution is core component which directly relates to the quality of video. The aim is to produce an efficient forensic video processing framework to assist the forensic crime scene investigation. This paper is aimed to perform video enhancement and noise reduction. The input video is processed using filtering techniques to reduce noise, artifacts and enhance quality. The real-time YouTube video url is passed as input to the model. First the video is pre-processed using steps. Initially in order to manage the issue of growing size of video due to upscaling we first aim at cropping the video down to certain time frame of interest. The cropped video is decomposed into n number of frames. QuadGRN is used to remove the background noise of frames. A ZeroNoise gate uses new MedAge filter with combined function s of average filter and median filter to refine video frames from noise. To normalize and have equal distribution of illumination EvenLight shedding module which functions based on clahe algorithm is used. A tailored kernel is selected to process the frame of interest by pel values. A convenient sequential deep learning model using CNN-GRN is built to process single frame input and to get single frame output using keras. We run this model using GPU to efficiently process the video.