Application Of V System In Texture Classification

As a common physical attributes, texture can be found everywhere in nature. The classification of the object by texture analysis on material has very important practical significance. In computer vision, texture classification has important research significance. In addition, texture analysis has been used in image retrieval and classification, road traffic areas, face recognition and other fields.To classify the scaled and rotated texture images correctly, this paper proposes a new algorithm for texture classification by combining Radon transform and the V-system. We firstly use the Radon transform to convert the image rotation into the image translation, and then apply the V-transform on the image obtained after Radon transform. The energies of the image on different levels under the V-system are expressed by performing a series of down sampling process due to the multi-wavelet characteristics of the V-system. These obtained energies are used as the texture feature description. The feature description method in this paper is robust to the image scaling and rotation because of the multi-resolution characteristics of the V-system and elimination of rotation by applying Radon transform. Results of the experiments conducted on the standard texture datasets show that the proposed algorithm provides superior performance.Removing the process of down sampling is to reduce the computational complexity. We firstly use the Logpolar transform to convert the image rotation into the image translation, and then multiply by the V-matrix on the image obtained after Logpolar transform. After normalization, the8×8matrix of the low-frequency part are used as the texture feature description and classification based on the vector machine. Results of the experiments conducted on the standard texture datasets show that the proposed algorithm provides superior performance.According to the characteristic of texture image local similarity, we try to use the new method to extract feature vector. We use the down sampling decomposition on the image obtained after V-transform. Local energies are used as the texture feature description and measurement standard is SKLD distance. Classification based on the nearest neighbors. Results of the experiments conducted on the standard texture datasets show that the proposed algorithm provides superior performance.