Lens clustering method based on information bottleneck theory

Abstract

The invention relates to a lens clustering method based on an information bottleneck theory. The information bottleneck theory is used for the lens clustering. Every lens to be classified is initialized into one sort. Common information loss in a merging process is used as a measure for a category distance to implement the sort combination until all lenses are merged into one sort. The sort number corresponding to common information loss inflection points is used as the final sort number of the lens clustering. At the same time, in order to solve the problem of local optimal solution appearing in a hierarchical clustering, the classification result is regulated when combined with a sequence clustering method. The lens clustering method can apply the information bottleneck theory to the clustering method without estimating the number of the sort beforehand, thereby improving the precision and the recall ratio of the clustering results and avoiding the problem of local optical solution.

Description

technical field [0001] The invention belongs to the field of video content analysis and retrieval, and in particular relates to a method for clustering shots. Background technique [0002] With the development of network and multimedia technology, digital video is more and more widely present in people's living space. However, due to the large amount of video data, how to effectively organize and manage video data has become one of the main problems faced by the development of multimedia applications under the condition of limited storage capacity and network bandwidth. Therefore, related research on video structure and content analysis has received extensive attention. Related work includes shot boundary detection, keyframe extraction, shot clustering, scene analysis, and event detection. Among them, shot boundary detection is the first step in video content analysis. Its main function is to decompose the video into its basic unit-shot according to the continuity of the u...

AI Technical Summary

Problems solved by technology

The shared information loss gradually increases during the merging process until the inflection point appears; after the inflection point, the shared information loss increases significantly

The number of categories corresponding to the inflection point of shared information loss is used as the final category number of shot clustering. However, bottom-up hierarchical clustering tends to fall into a local optimal solution

Images