Performance analysis of different kernels of Support vector machine and self-organizing maps for classifying musical and non-musical personal video

Abstract

Objectives: This study aims at detecting the video with or without music. The present article will support the contextualization process of the videos. Method: Videos with musical piece are detected using temporal and spectral features. Its focus is also on the comparative study of two classifiers i.e. Self-organizing maps and Support Vector Machines. Findings: Music detection in audio track of video was mainly carried out for professional videos. The proposed work focuses on personalised videos which are very challenging due to limitations in the environmental conditions as well as procedures of recording. No work is found on the use of musical portion detection especially in the personal video recordings. The successful detection of music in audio track of videos is used for prioritization in the contextualization process. Accuracy of detecting musical video is 85% with combination of temporal and spectral features. The limitation in the accuracy is due to non-professional recordings of the personal videos. Application: With recent advancement in hardware technology, the high-end HD cameras are available in every handheld device. The video capture is significantly increased especially in the personal videos. The use of images and videos are becoming essential part of everyone’s routine from students for studies to elderly individuals for safety and entertainment. It is forecasted that the 80% of internet traffic will be occupied by the videos. Hence video analysis has become very crucial to detect unwanted video portions to avoid the transportation of these videos. It is a vital task to design a model to detect the videos having musical portion for prioritization for video editing and contextualization process. Keywords: Audio-visual Stream Analysis, Detection of Music Video. Multimedia Signal Processing