The R&D Of Color Image Feature Extraction And Recognition

Since the 1980s, machine vision technology will become a very active field of research.But color vision solution development has been slow. However, with the microprocessorprocessing ability enhancement and the machine vision application trade continues to expand,there appeared must use color visual industrial applications, such as food quality grading,packaging detection, prints quality test, tablets color recognition application.In the HSI paper using the color information of color smart camera as the main feature,R&D some kinds of color visual function modules and solutions, in order to adapt to theapplication of the color vision in industrial automation.According to industrial application of color visual function modules, basing on thetraditional automatic white balance algorithm, proposed Bayer gray world-retinex whitebalance algorithm to process the Bayer pattern image data directly. Implemented colorextraction function module in HSV color space. Based the color extraction function module,and transformed the pixel data of the non-extraction zone into gray via HSV Color space,realizing the shade processing function module, meet the requirement of neutron pixelposition detection in color visual application.For the demand of color vision solution application in the classification solution,proposed the floor hierarchical classification based on the color characteristics. To classify theunknown floor samples, the first use of the shortest distance classification to determine thegeneral category of the sample, then use the improved K-nearest neighbor classification todetermine the specific categories.The article completed the color visual function modules and solution, from the view ofthe project R&D. By test: the Bayer Gray World-Retinex white balance algorithm caneffectively reduce the color temperature with a low data computation and process the640×480 original image data in the smart camera with 600MHz within 34 ms. Colorextraction function module has the efficient regional extraction performance, in smart camerato process the 640×480 non-compression color image needing about 50 ms. The classifiedaccuracy of floor hierarchical classification program is 95.6%, and the program in the PCchance done time for 390 ms classification. Fully verified the color visual function modulesand solution have met the requirement of the practical industrial application about practicaland of real-time.