Texture Segmentation Method Based on Multivariate New Symmetric Mixture Model and Log-DCT Features Derived from Local Binary Patterns

In the proposed methodology, explore the superiority of different texture regions in an image along with log DCT and LBP. The image entirely metamorphoses into a local binary pattern domain using this technique. The ensuing step is to generate non-overlapping blocks based on the LBP image. The DCT coefficients are obtained in a zigzag pattern from each subsequent block. The EM technique is used to evaluate the model parameters, under the assumption that the feature vectors follow a new multivariate symmetric mixture model. The model parameters are initialized using the moment estimation technique and the hierarchical clustering approach. A Bayesian framework and the maximum likelihood method were used to develop the texture segmentation algorithm. In the Brodatz set's database, randomly selected images were used to implement the proposed algorithm using performance metrics like GCE, PRI, and VOI. Considering the performance metrics, this technique is better than the current texture segmentation algorithms. The segmentation accuracy is quantified via metrics like GCE, PRI, VOI, accuracy, precision, recall, F-measure, enhanced correctness, and enhanced model performance. Comparative studies show that integrating both LBP and log-DCT increases reliability and precision in texture segmentation.