Texture Segmentation Using Threshold Based Micro Texture Unit Approach

摘要

Image segmentation is one of the crucial steps for image analysis, interpretation and recognition. The present paper proposes a new "threshold based micro texture unit (TMTU)" segmentation approach for effective image segmentation. The texture unit approach characterizes the local information for a centre pixel and its neighborhood by deriving small units called texture units. In the basic approach the texture elements are treated as ternary without any threshold and TU's are evaluated on the 3×3 neighborhood. The basic approach is derived on eight neighboring pixels and the range of this TU is very high. The proposed TMTU derived ternary (base 3) and quinary (base 5) texture elements using threshold and partitioned the 3×3 neighborhood into 6 micro texture units. These six micro texture units corresponding to 3 rows and 3 columns and each consists of three texture elements only. The proposed TMTU segmentation scheme is derived on uniform local binary pattern (ULBP) that divides the image in uniform local regions. The proposed TMTU is experimented with four different databases i.e. WANG, Oxford flower, Brodatz, Indian facial images, standard images and compared with the existing algorithms.