Genre based video retrieval using similarity function between feature vectors

Genre based video retrieval approach is proposed in this paper. First, video clip of representing a specific genre is segmented into diminutive parts. Then, six low-level features such as Saturation ratio, Motion ratio, Bright ratio, Average brightness, Average edge ratio & Flash ratio are extracted from each part of video. Proposed algorithm generates feature vectors of each corresponding part. These feature vectors are then represented in the form of chromosomes. After indexing all the chromosomes into the database, a genetic crossover operation is applied to those reference chromosomes out of all database chromosomes which have specific genre as per user requirement of retrieval. Subsequently, new after chromosomes crossover process, all the other chromosomes in the database are compared based on a similarity function to obtain the most analogous chromosomes as solutions of the subsequent generation. By this process, a new portion of the video which contains the specific genre is retrieved. In order to show the validity of the proposed method on seven categories described as action, war, car race, happiness, Sadness, song and conversation used as genre for empirical experiments. Happiness scene in the video contains bright tones and their motion dynamics was relatively low. So Average Brightness and Motion Ratio features are useful in the process to retrieve the happiness scene. Similarly, in most of the scenes like war, action, and car race contain dark tones and high motion. To evaluate and analyze Saturation ratio, Motion ratio and flash ratio are used in the retrieval process of these genres. At the end, simulation results generated by applying this algorithm on seven videos based on different genres, which reflect more about this idea.