Reinforcement Learning for Enhanced Content Based Video Frame Retrieval System in Low Resolution Videos

Over the past two decades, evolutional advancement took place in the field of computer, communication and multimedia technologies which led to the production of massive video data production and large image repositories. This research focuses on the video frame retrieval using popular feature extraction method of computer vision called Speeded Up Robust Features (SURF) and Reinforcement Learning (RL) based Point of Interest (POI) calculation and frame reduction. The proposed method uses the concept of Content-Based-Image-Retrieval (CBIR) which retrieves similar images from the collection of large image databases. In video processing, a challenging task is to remove redundant frames as most of the videos have 30 Frames per Second (FPS). This research study consists of the removal of redundant frames and retrieving of similar frames from remaining non-redundant frames based on a query image. Validation for the proposed scheme has been achieved by simulating the system for a standard dataset as well as videos captured ourselves. The results depict the efficacy of the system and provide a platform for benchmarking.