A quick search method for audio and video signals based on histogram pruning

Abstract

Abstract-This paper proposes a quick method of similaritybased signal searching to detect and locate a specific audio or video signal given as a query in a stored long audio or video signal. With existing techniques, similarity-based searching may become impractical in terms of computing time in the case of searching through long-running (several-days' worth of) signals. The proposed algorithm, which is referred to as time-series active search, offers significantly faster search with sufficient accuracy. The key to the acceleration is an effective pruning algorithm introduced in the histogram matching stage. Through the pruning, the actual number of matching calculations can be reduced by 200 to 500 times compared with exhaustive search while guaranteeing exactly the same search result. Experiments show that the proposed method can correctly detect and locate a 15-s signal in a 48-h recording of TV broadcasts within 1 s, once the feature vectors are calculated and quantized. As extentions of the basic algorithm, efficient AND/OR search methods for searching for multiple query signals and a feature dithering method for coping with signal distortion are also discussed.

An Audio Retrieval Algorithm Based on Audio Shot and Inverted Index

“…We compare the proposed algorithm with a conventional method called the time-series active search (TAS) [2], which models the distribution of feature vectors with histograms and accelerates the search with dynamic skip width.…”

“…We propose the concept of audio shot for audio segmentation. Besides, traditional methods [2] apply a sliding window to search for matching segments, which is very time-consuming. To filter out the irrelative parts efficiently, we apply the inverted file that is popular in text retrieval to audio content indexing.…”

Challenges and techniques for effective and efficient similarity search in large video databases

“…Matching of similar videos is often translated into searches among these feature vectors [14,18,20,16,35,8,25]. The number of feature vectors depends on the length of video.…”

“…The primary difference between these two scenarios of similarity queries is that, for the former, the clips for search have already been segmented and are always ready for similarity ranking [35,8,25], the latter is a typical subsequence matching problem [14,18,20,16] (which is conceptually analogous to subsequence matching in time series [10]). Because the boundary, and even the length of target subsequence are not available at beginning, choosing which subsequences for evaluating similarities is not pre-known.…”

“…In multimedia, most existing research efforts are on contentbased video copy detection [14,18,20,16,35], which is regarded as a complementary mechanism of video watermarking. Their primary focus is to detect any video sharing exactly the same original source with query, but can be altered with some global transformation, such as different formats, different resolutions, changing brightness, changing contrast, changing saturation, cropping, overlaying a logo, etc.…”