Automatic Detection of Colour Objects Using Spatially Weighted GVF

摘要

Automatic detection of visual objects in digital image still presents one of the great challenges in computer vision applications. In this paper, we present an enhancement of the Colour Edge Co-occurrence Histogram (CECH) and Colour Edge Gradient Co-occurrence Histogram (C E G C H) for detecting objects in unconstrained colour images. This paper describes an automatic approach that performs visual object detection by combining the strength of processing the colour edges along with spatially weighted gradient vector flow (SWGVF) of intensity image. We introduce SWGVF for constructing an accurate representation of object edges in the image, as compared to the simple colour difference classification which is done in CECH and colour edge detection based on the R-ordering method using vector order statistics, defined in CEGCH. Our proposed method is Colour Edge Spatially Weighted Gradient Vector Flow (CESWGVF) and is based on vector force field of pixels and histogram of the colour edges in image. We employ a perceptual colour quantization compared to CIE colour quantization scheme used in CECH and HSV (Hue, Saturation, Value) colour space quantization in CEGCH. Experimental results show that the CESWGVF detects various colour objects with comparable accuracy as that of CECH and CEGCH and with a good detection rate compared to CECH and CEGCH. Potential applications include real time object detection in video and content based image retrieval.