Global Windowing Based Thepade's Sorted N-Ary Block Truncation Coding (TSnBTC) for Content Based Video Retrieval with Various Similarity Measures

Abstract

In this modern digital era, the Content Based Video Retrieval (CBVR) utilizes the video contents for representation, indexing and retrieval applications. The paper attempts Content Based Video Retrieval with TSnBTC (Thepade's Sorted n-ary Block Truncation Coding) color feature extraction technique and windowing of RGB planes into number of different partitions, such as windowing in <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$1\times 1,2\times 2,3\times 3,4\times 4$</tex> and then TSnBTC is applied on these partitions. This Paper analyses the effect of 11 different similarity measures on accuracy of video retrieval. The similarity measures used are distances like Euclidean, Chebychev, City Block Metric, Hamming, Mean Squared Error, Soergel, Sorensen, Canberra, Kulczynsk, Cosine and Jaccard across 3 families such as LP Minkowski family, L1 family and Inner Product family in which Sorensen distance gives higher accuracy immediately next higher accuracy is given by City Block Metric distance. The process of feature vector extraction by using 20 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">th</sup> frequency frame of videos. The test bed of 500 videos of different 10 categories of video sets is used for experimental appraise of proposed TSnBTC based video retrieval method. Each testbed category consist of 50 videos. The average video retrieval accuracy is computed for each of the partitioning variations of proposed technique. Through experimentation it has been found that TSPBTC (Pentanary), TSSBTC (Septanary), TSOBTC (Octanary) gives better video retrieval accuracy as compared to TSTBTC (Ternary) and TSQBTC (Quarternary) and partitioning (global windowing) improves the accuracy of content based video retrieval.