Texture based color segmentation for infrared river ice images using K-means clustering

The problem of texture Segmentation involves subdividing an image into differently textured regions. Gabor filters produce outputs which are notably distinct for the different textured regions. Detecting the discontinuity in the filters output and their statistical properties help in segmenting and classifying a given image with different texture regions. Feature calculation for every pixel in the image reduces the computational cost for color based segmentation. In this proposed method, firstly image texture segmentation is performed by using Gabor filter and the result obtained from Gabor filter differentiates the textures with a variety of colors as different textures, and these similar colors are extracted as different images by color-based segmentation with K-means clustering and finally the features are extracted by using first and second order statistical methods. The features extracted from first and second order statistical methods are given to PNN classifier. Using this methodology, it is found that texture and color segmentation followed by gray level co-occurrence matrix feature extraction method gives higher accuracy rate of 95.5% when compared with other feature extraction methods.