Duplicate Video and Object Detection by Video Key Frame Using F-SIFT

Abstract

To explain and detect different features in images scale-invariant feature transform can be used effectively. From starting, a set of reference images SIFT important points of objects are extracted and stored in a database. An object in a new image can be recognized by individually balancing each feature from the new image to this database and then finding features for candidate matching. As a valuable local SIFT can be utilize as a solution point descriptor for its invariance to lighting, scale, and rotation changes in images. Since SIFT is not flip invariant, flip invariant SIFT is planned. These F-SIFT is established to identify large scale duplicate videos, object finding as well as recognition. It requires to take out all the frames from query video and videos in dataset for similarity matching, time complexity of f-SIFT is more, So to remove such limitation we have projected dual threshold technique. Our method will eliminate redundant video frames by applying auto dual threshold method. So there will be no necessity to execute the extraction of features and matching of sequence with all video frames. Unnecessary frames are detached by making segments of video. Only the key frames are extracted for matching purposes. Here we are using two thresholds. The first is for identifying direct changes of visual information of extracting frames and other second for detecting usual changes of visual information of extracting frames. Threshold values are decided as per the information of the video. This system, extracting total three frames like first frame, last frame and key frame from video segment. By using the average feature value of all the frames in the segment, key frames are decided. For similar propose a key frame is used and remaining two frames are used to detect the segment location.