Texture recognition by using a non-linear kernel
International Journal of Computer Applications in Technology
| Hi-index | 0.00 |
Direction parameter θ is one of the important parameters of GLCM (gray level co-occurrence matrix). A fixed angle (such as θ=45°) or the average of themeasurements in four direction (θ=0°,45°,90°,135°) were usually used in calculating GLCM. However, these methods are just empiristic idea, lacking of theoretical support. In fact, the above-mentioned idea are usually failed to describe the image texture, especially for those texture images with strong directional characteristics. In this paper we propose a new method of choosing the maindirection of texture image by calculating the correlation of GLCMs of different direction. Through selecting the texture characteristics value of main direction, combine with the average of the measurements in other three directions, a set of characteristics which includes more texture information and rotation invariance were extracted. The experiments on Brodatz and Outex texture database show that the characteristic set we selected is more discriminate and more accurate.