Abstract
Image recognition systems widely use grayscale images to perform analysis. One of the most important function in image processing tend to be edge detection and its different forms (i.e. object edge, reflectance edge, illumination edge, specular edge, and occlusion edge) [1]. The simplicity of a grayscale image, which only contains a single channel, makes it a preferred option in image recognition systems. Converting color image to a grayscale image may be a sufficient input data for these functions, but if the need exists to differentiate colors within an image, grayscale would not hold sufficient information to be used as input data. Therefore, a multi-channel format such as the RGB or HSV color space needs to be considered. A new optimization algorithm that can be applied directly onto a color image for segmenting features based on their respective colors is created and utilized as an alternative to converting color images to grayscale before analyzing. The Color Image Segmentation Algorithm (CISA) is designed to be applicable in the inspection or reverse engineering process of electronic components. Therefore, color image data used as inputs to CISA are obtained from electronic components such as layers of printed circuit board (PCB). Each pixel values of a given color image is first plotted in a 3D coordinate system utilizing the RGB color channels. The RGB coordinate plot is used to correlate and reassign pixel values to user input RGB values to perform the color image segmentation. The process of using clustering methods to reduce initial user inputs is also explored and could be a key improvement to CISA in future updates. Results obtained by inputting a PCB layer image through the color image segmentation algorithm show clearly defined segmentation of copper routing from the FR-4 material. In comparison, a simple color-to-grayscale conversion generates an image with a contrast level that makes the copper routing very difficult to segment from the FR-4 material.