Development of Feature Descriptor for Texture Classification

Image classification strategies based on surfaces play an imperative part in different computer vision applications. The challenge exists in texture classification are rotation, illumination, scale and orientation changes. Our proposed method uses efficient features for texture classification that overcomes all the above challenges. The steps involved in this algorithm are: The basis filters were calculated in multiple scales, by using the first and second-order Gaussian directional derivatives. These basis filters are convolved with input image to compute max-min responses. By applying linear and non-linear administrators on max-min reactions, the highlights have been extricated. The 2D-DWT is applied to the same input image. The Local Binary Pattern features were extracted for the sub-bands of 2D-DWT applied image. Scalar quantization is grasped to quantize these texture features into surface codes, as these feature set are high-dimensional which are unseemly to concatenate all highlights over a entirety picture. The cross - scale coding has been performed over these quantized codes for a compact representation of the histogram. This compact histogram representation was used as an image feature representation for the classification process. For classification KNN classifier is used. The performance of the algorithm is validated by using OUTEX-10 and OUTEX-14 datasets.