Research On Animal Image Detection Based On Generalized Hough Transform

Animal image detection is the major difficulty in the field of object recognitionalgorithms. Due to parts of animal body possess high degree of flexibility, when themotion state changes its posture also changes constantly, which causes difficulties indetecting. According to above problem, this paper presents a detection model based onthe Generalized Hough Transform, by which integrating local shape fragments featuresare extracted from animal for achieving the target of animal detection. This paperfocuses on the issues of the edge extraction methods, detection model building, shapefragment evaluation and detection algorithm integration and so on.Firstly, analyzing threshold selection problem of the edge extraction process in thepre-processing, by which building the Canny hierarchical structure detects edges, anddesigning the connected algorithm based on connected domain approach complementimportant shape fragments，while reducing the interference of the background noise inedge image layers.Secondly, the algorithm votes for the shape fragments of the back and buttockseparately based on Generalized Hough Transform approach, and positions the localextrema location using clustering algorithm in Hough voting space, and then evaluatesthe similarity among the shape fragments which are obtained by the local extremalocation, by which selecting shape fragments that satisfy the conditions of similarity.According to the relative positional relationship, the algorithm defines the leg areabased on the locations of back and buttock which is satisfied with above conditions, anddetects the line segments of legs. On this basis, counting the number of exact legfragments achieves probabilistic evaluation for leg area.Finally, combining evaluation of back, buttock and leg area with relative positionalrelationship between back and buttock designs detection algorithm, the detection resultis the area corresponding to the maximum value of the calculated result. By detectingINRIA Horses Database verify the validity of the algorithm.