Research OnContour Grouping Algorithm Based On Spatio-Temporal Cues

The purpose of contour grouping is to extract an independent object boundary from the input image. It is also a perceptual organization process which treatsboundaryfragmentsas the grouping classes. Contours can well describe the geometrical features and topological characteristics of the target and the contour descriptions are veryconcise,so contour grouping is of great sigificance for the subsequent high-level vision tasks, such as object recognition, scene analysis, content-based image retrieval.This paper is based on Gestalt grouping principles. The image sequences are treated as the input. A novel contour grouping algorithm based on spatial and temporal information is presented. The main works of this paper are as follows:1、Study edge detection algorithms. According to the characteristics of contour grouping, a boundary fragment grouping extraction algorithm based on gPb detectoris proposed.2、The extracting methods of spatio-temporal cues are studied. A motion feature extraction algorithm based on the LDOF optical flow methodis proposed. And a space feature extraction algorithm based on the color and texture descriptionis presented.3、The manifold learning methods are discussed. Then, compare non-linear dimensionality reduction with linear dimensionality reduction, anda spatio-temporal similarity algorithm based on manifold learning is presented.4、Study clustering algorithms, treat thespatio-temporalsimilarity as clustering information. At last, this paper propose a contour grouping algorithm based onadaptive spectral clustering.The novel algorithm is tested in Stein and BJTU_motion dataset and achieves satisfactory results. The grouping efficiency of our method is72.52%. The research content of this paper is the basis of many senior visual tasks.