Planar Object Recognition Based On Characteristic-Ratio

A planar objects recognition method under sever projective projection based on Characteristic-Ratio resistant to perspective deformation is proposed in order to prove the efficiency of projection-invariants in object recognition. First, contour of the image is extracted and a serious of straight lines cross the image is constructed. Then, the image is represented by a sequence of characteristic ratio spectra. Finally, Dynamic Time Warping algorithm is employed to compare the similarity of spectra and to find the pixel level correspondence between two images.First, we introduce the theory of perspective transformation and the concept of projection invariants. Starting from the model of center projection, this paper introduces the basic formula of perspective transformation and the transformation matrix, proves that the projective transformation is a distorted transform which is a combination of a series of perspective transformation form, and explains that in real life we see objects from different angle observation will cause the projective distortion. Then leads to homogeneous coordinates as the theoretical basis for the following characteristic ratio, and then introduces the definition of cross ratio and character ratio.Second, followed by the introduction of the definition of convex hull and several commonly used algorithms, and then the convex hull points are connected as the outline of the object, then we get uniform sampling in the outline of the object every same distance, and see sampling points as feature points. Based on the analysis of a specific sampling point, we will see that deformation caused by the projective transformation of plane target can be transformed into the characteristic spectra of stretching and stretching transformation spectrum, and simplify the object recognition.Finally, the final distance between two image obtained by two layers of DTW algorithm. The nearest neighbor recognition strategy is used, the template image with minimum distance from the test image as the recognition results. Experiments show that, this algorithm not only resistant to server projective transformation, but also improve the recognition rate on the objects with very complex or simple structure.