Study on unsupervised color texture image segmentation techniques using morphological watershed algorithm

Abstract

Image visualization, object based image compression, object detection and recognition, content based image retrieval are highly dependent on the segmentation results. Five different techniques are investigated. Two unsupervised image segmentation methods have been proposed to segment color images, with or without texture regions. The proposed methods are analyzed with respect to the performance parameters such as Borsottiand#8223;s Evaluation function Q which is an empirical goodness evaluation metric and execution time. A low value on the evaluation function indicates better segmentation. The proposed technique takes lesser time in partitioning the images but the perceptual segmentation quality is not good as it produces over segmented regions. The second proposed segmentation algorithm is for segmenting the natural scenes with two low level features, color and texture. The third method improves the perceptual segmentation quality. Adaptive mean-shift clustering is used to finish color classification Next Conditional filtering is applied over the image regions having non uniform texture characteristics. Two methods have been proposed to reduce computational time by processing blocks instead of individual pixels. The first method considers the segmentation problem based on color or texture information in a non parametric framework. The second proposed methodology is for the segmentation of color images using intensity, position, color and texture features to facilitate the formation of regions corresponding to the real objects. The results clearly demonstrate the superiority of the unsupervised segmentation scheme using K Means connectivity constraint algorithm with marker controlled watershed as initial clustering stage in producing accurate boundaries with high perceptual segmentation quality. An image retrieval methodology for search in large collection of heterogeneous images using the fully automatic segmentation algorithm is also presented. This work can be extended to detect the moving object in a video clip. newline newline