Research On Extraction Of Texture And Quantitative Color Information For Flat-granular Object

To improve the situation in which precisely separation for flat-granular objects, like watermelon seeds, is still being performed by hand according to the latest investigation, extraction and data processing of important information for texture and color feature contained in melon-seed color image has been deeply analyzed and studied in combination with image processing and mathematical statistics principle.Take watermelon seeds from Huangdai, Suzhou city as research objects, three detailed jobs are as follows:(1) Four methods describing texture feature of object, like statistical methods, structural methods, signal processing methods and fractals methods, have been analyzed and compared to summarize the distribution of texture for watermelon seeds by using gray co-occurrence matrices to extract five descriptors based on experimental results;(2) A BP propagation classifier has been designed to quantize full-color image and extract eleven important texture information contained in color image of watermelon seeds through the results of analysis and selecting one of the best training functions from four traditional training functions in neural network;(3) Twenty-two descriptors including texture, color and geometric features of watermelon seeds have been summarized to customize specifications for application on separating watermelon seeds via computer and complete the analysis, design and development tasks for standard database based on mathematical statistics principle.