Texture segmentation algorithm using multivariate new symmetric mixture model and DCT coefficents under logarithmic domain

Texture analysis plays a crucial role in image analysis and processing, particularly in texture segmentation, which has increased important consideration due to its applications in computer vision and automated scene recognition. Numerous texture segmentation techniques have been proposed, often based on the hypothesis that the texture features of an image section follow a multivariate new symmetric mixture model. However, the limitations of univariate new symmetric models, which are inherently mesokurtic, restrict their ability to accurately characterize textures in diverse image regions. To report this, this training familiarizes and assesses a texture dissection procedure that undertakes the texture landscapes of a whole image are characterized by a discrete new balanced combination technique. This model encompasses a variety of distributions, including leptokurtic, platykurtic, and mesokurtic as special cases. The algorithm’s limitations are projected using the expectation-maximization (EM) procedure, while segmentation is achieved using a Bayesian framework and maximum likelihood estimation. The proposed method is validated using sample images from the Brodatz texture database and assessed using established segmentation quality metrics. Comparative analysis discloses that the projected system outclasses current methods based on the multivariate new symmetric combination model in terms of segmentation quality.