Abstract:

Segmentation is the process of splitting the image into regions, which are visually distinct and uniform with respect to some property, such as texture or color. In order to properly segment texture images, texture watershed transformation can be applied. Though it gives a better result than the raw watershed, it will not give a complete solution to the over segmentation. Hence it needs a post-processing. Here, a two stage method for
segmenting both textured and non-textured images is proposed. First stage is the texture watershed in which texture gradient is computed from the detailed coefficients of the dual-tree complex wavelet transform, for gray scale images. Texture gradient is computed for the three planes of the RGB color images. These gradients are combined to get a single texture gradient which in-turn combined with the intensity gradient to get the final gradient, to which watershed transformation is applied to produce an initial segmentation. Second stage is the weighted mean cut clustering technique, which recursively partition the graphical representation of the connected regions of the pre-segmented image, till the partitioned graph represents only the homogeneous regions. The proposed algorithm produces effective segmentation for both grayscale and color images, with or without texture regions. The performance of the algorithm is measured quantitatively with the help of  segmentation benchmark..

Keywords: Color image segmentation, complex wavelet, texture, textured watershed, clustering.

BibTex:

@ARTICLE{P1150619002,

AUTHOR = {S. Kothainachiar and R.S.D. Wahita Banu and A. Saravanan},

TITLE = {Unsupervised Morphological Segmentation for Textured and Non-Textured images},

JOURNAL ={ICGST International Journal on Graphics, Vision and Image Processing},

}

(Full Paper 905KB)