Separating a Colour Texture from an Arbitrary Background

Partially supervised segmentation, that is, segmentation with always incomplete training data has many practical applications in image analysis and retrieval. This paper proposes a new algorithm for finding regions of a single texture in an arbitrary colour image. The texture is specified by a given training sample. The algorithm exploits colour space vector quantization, colour thresholding, and similarity between characteristic grey level cooccurrence histograms over a moving window around each pixel and over the whole training sample. Experiments show this algorithm effectively finds various homogeneous textures in complex backgrounds.

• Type: Conference
• Year: 2003
• Where: DICTA
• Authors: Linjiang Yu, Georgy L. Gimel'farb
• Keywords: Algorithm Exploits Colour, Arbitrary Colour Image, DICTA 2003, DICTA 2007, Training Sample