Research On Image Matching Algorithm For Train Fault Based On Geometric And Shape Features

Computer image processing is a rapidly developed emerging discipline, widelyused in various fields. It is based on human vision and it strengthens the human visualinformation in the image information for further processing, and for more refinedinformation. In practical applications, shape, color, texture and other features arecommonly used in image recognition, particularly the shape feature as the main basisfor retrieval, has the superior characteristics that color texture can not be compared, andthen more and more image retrieval algorithms based on shape features have beenexcavated. However, through the study of various types of shape matching methods,various algorithms have to adapt to the application background, so far, there are not anyshape matching method wich can be used to all of image matching applications withoutany changes.In this paper, the object is one of the faults in Trouble of moving Freight carDetection System (TFDS), truncated plug door handle identification. TFDS system isdeveloped for vehicle safe. Affected by train speed, the natural environment and otherconditions, there is much noise in the collected images. First, the relevant background,fault types and features are described and analyzed. Two image recognition methodbased on geometric features and shape descriptors are presented. Both focus on imageedge information. Therein, the method based on geometric features is easier as bringingout the result by external rectangle. Otherwise, the method based on shape features mustuniformly sample the contour points in order to find the centroid of the image. Then, theskeleton function is defined through taking advantage of the height function to describeshapes, and find out the shape descriptor. Finally retrieval purpose is achieved throughtemplate matching on shape descriptor. The shape descriptor is an N-dimensional vector,N is the number of sample points. Wherein, to form a connection from the centroid toeach sample point as a reference axis, in turn to calculate distance from each samplepoint to the corresponding reference axis, each distance value is a element in theN-dimensional vector. Using skeleton function to define descriptor is an innovation. Notonly it reveals well the relationship between feature points and contour, but also solvestranslation, scaling and rotation problems by determining the starting point based onhalf axle angle.Compare with grayscale template matching method, the two methods mentionedabove not only compute faster but also accurate better. Between these two methods, theone based on shape descriptors performances better in anti-interference ability. Thus,expressing the contour by shape descriptors for similarity matching achieves theadvantages of high speed, high precision and practice, it is more suitable for compleximage recognition than the methods on geometric features.