Segmentation of Restored Colour Texture Images Using MRF with Cellular Automata and Colour Block Matching 3D Algorithms

Abstract Role of texture segmentation is vital due to its diversified applications in image analysis and understanding. Noise is an integral part of measurements and affects the performance of segmentation algorithms. Recently, researchers have initiated the work on noisy textures to address automatic quality evaluation of fruits and vegetables, diagnosis of thyroid cancer images, recognition of facial expressions, texture-based image retrieval and texture analysis, after restoring the noisy images. With these motivations, benchmark images from the Prague and Brodatz texture dataset are tainted with salt pepper and Gaussian noise and restored using recent cutting edge performance algorithms. A novel Markov Random Field based approach using a custom Median filter is developed for segmentation of restored texture images. A maximum improvement of 16% segmentation accuracy for salt pepper noise with 70% noise density and 15.1% accuracy for Gaussian noise with variance 50 has been achieved over recent approaches on Brodatz textures. An improvement of 8% accuracy for Gaussian noise with variance 50 and 4.47% accuracy improvement for salt pepper with 70% noise density is achieved on Prague texture benchmark images. The proposed approach can be applied for satellite and medical image analysis, industrial applications, remote sensing.