Object Recognition Based On Shape Context

Object recognition is a basic research in the field of computer vision. Effectiveobject recognition algorithms in the intelligent video surveillance, image analysis,object recognition, image retrieval, and other areas of the application play veryimportant roles. So far, people have many great achievements in object recognitiontechnology, but there are still many difficulties need to be solved.Shape Context algorithm is studied in this thesis. This thesis describes the basicideas and implementation methods of Shape Context algorithm and Objectrecognition method to identify and match. For Shape Context requirements of theinput image, discuss some basic image preprocessing methods and edge enhancementalgorithm and extraction methods of Canny algorithm. A set of core processes ShapeContext object recognition for image processing and recognition is designed in thisthesis. With this process, an experiment and verification software is developed andimplemented using C++language.In the implementation of the algorithm and software tests, we found theshortcomings of the traditional Shape Context:(1) First need to calculate the centroidto determine the number of extracted feature points, this process is time-consumingand need a large amount of calculation;(2) The contour feature points extracted maynot be well representing the shape of the object.This thesis proposed an improvedmethod for the two main aspects. First we treat the tested object image into a binaryimage, and use the Canny edge detection algorithm to get the edge, and then obtainouter contour with the contour tracking method. according to the number of pointsobtained outside the contour of the object set feature extraction point distance, thefeature point can be better expressed contours of objects, and the feature pointinformation is stored in the document, when we calculate the distance between twoobjects shape feature points, we can call it directly. with this we can not only savemore time, but also can be a good characterization of the shape of the two objectsdegree. In addition, this article uses TPS (thin plate spline) transformation model to simulate the deformation of the tested object to the database objects, calculate theminimum bending energy, it and the two object Shape Context distance form theshape matching distance cost, it is a good characterization of the difference betweentwo objects shape. After experimental verification, Shape Context algorithm toidentify the object shape soon, and the effect is better.