Blurring of inappropriate scenes in a video using image processing

Abstract

Inappropriate scenes such as bloody scenes, nudity, gore, drugs, weapons etc. are considered inappropriate especially in a developing Muslim country like Bangladesh. In our country, such scenes are very discouraged for children, old people or heart patients. Thus, keeping these in mind we propose a model where these inappropriate scenes will be detected and blurred from any video stream. Moreover, the model also shows percentage of explicitly in an input video file. As a result, people playing the video will know beforehand whether they would want to watch it or not and parents will have a greater control on what their children are watching. For nudity detection, there will be fragmented human figures that will be extracted and then the fragments will be compared against a database to decide whether nudity is involved or not. If it is involved and exceeds a predetermined threshold, then the video will be considered as pornography that many people may not prefer to watch. Similarly, the extracted figures of objects or gore scenes will be compared against a database to know the percentage of inappropriate scene in a video. Bringing both the nudity and goriness under one roof, we named the term explicit and if a video is explicit, the user will have the option of knowing it from beforehand and blur out any portion from the video automatically by using our model. The accuracy of our model is 93% and the algorithm we have used in this paper is CNN.