Object-Detection In Film

Object detection in movie by image processing

Abstract: Our aim is to detect the object in movie through image processing. If we generate the color blob image from our movie’s object and that image match from image database then our object is easily recognized. On the other hand running object is difficult to recognize the object in movie because if we consider about the sift for detecting the object in movie and that particular object are not so well detected object in movie. So paper we really want to create a generative model of that particular object of that movie and recognized through image processing.

Introduction: Today’s world have different types of approach of image processing. We really want to create system where we can easily

Everingham et al. use scripts and subtitles to learn the association between character names and faces in television drama using a frontal face detector. Sivic et al. demonstrate a much improved coverage (recall) by using profile views. On their experiment they reported that 42% of actor appearances are frontal 21% profile and 37% are actors facing away from the camera. Following their observation they says that, as a first step future work on increasing coverage should go behind the use of face detection. Generic methods for detecting upper-body or full-body actors have been proposed by Dalal et al., Eichner and Ferrari and Felzenswalb et al. They show that given the movable nature of the video, people can tracked only 50% in this approaches. Our proposed model of actors appearance related to recent work in object detection and recognition. we use a training set with front views, back views and side views of actors as shown in Figure

We want to detect the location first by search space reduction. KNN search is very effective for finding the motion of the actor and recognized its patterns. We applied the novel clustering algorithm to analysis the independent view framework of actor model. Head and shoulders of the actor are representation of color blobs where each blob is represented in a nine dimensional space. This nine dimensional space is combined with size, color, shape and position which is relative to the actor's co-ordinate system. It is based on a frequency term. In the first stage we reduce the search space using the k- nearest neighbours. This k- nearest neighbours are corresponding to the actor by just using the appearance of the model blobs. In the second stage, getting the best localization of the actor of all size, we used the sliding window search. Then we can get detection windows and actor names that maximize the posterior likelihood of each frame of the