| Image mosaic is the key to acquire the surface information of fruits. It is based on image matching. In this study, Red Delicious was used as the object to research the image matching. The main contents are as follows:(1)Construction of image acquisition equipment. First step is to analysis the demand of image acquisition equipment. Then, build the illumination system. In this system, LED was used with diffuse illumination method. Results showed that images obtained by the system had uniform illumination on the fruit surfaces, and the standard deviation of value R、G、B were less than8. Thirdly, the image acquisition equipment was constructed including illumination system, imaging devices, rotary displacement, and image acquisition software. Finally,34%area which was not close to the edge in the overlap portions of original images was selected as the actual matching image by the simulation and experimental verification.(2)An image matching method for surface images of Red Delicious was proposed, which was based on SIFT (Scale-invariant feature transform) algorithm and the method with dynamic coefficients of each component. Firstly, different gray-scale images could be obtained by changing proportions of three single-component (R, G, B), then was to make a matching for each kinds of gray-scale images based on SIFT. The results showed that the method had a success rate of higher than94%, while the proportion of mismatching points was about24%. And it cost more than5minutes for each matching calculation. Secondly, a culling algorithm was proposed to cull the mismatching points. It was based on the geometric position between the matching points and their adjacent keypoints and matching points. The results showed that more than95%of the mismatching points were culled successfully, and once matching cost0.016second in average.(3)S-SIFT algorithm (Simplified-SIFT) was proposed. It canceled the scale-space construction and rotation invariant of SIFT algorithm. The results showed that the time to detect keypoints by S-SIFT algorithm was reduced by56.3%compared with that by SIFT algorithm. For the image with resolution of0.146and 0.185mm/pixel, the success rates of mosaicing were both more than99%. The time of matching calculation was14.5%of that by SIFT algorithm in the2octaves and2intervals images.(4) The image matching algorithm based on spots extracting and vectors which between points and its adjacent points were proposed. Firstly, considering different conditions of fruit surface, this study presented to apply the Gauss difference image detection, Harris corner detection and Canny edge detection to extract spots. The next step was to match the spots. It was based on the vectors between the spots and their nearest or second nearest spots. The spots extracting method and image matching method were validated by the experiment. The results showed that the matching method could search the matching point stably with success rate was higher than99%for images of different resolution. This study of culling the mismatching points presented to judge the vectors between the matching points and their nearest or second nearest spots. The results showed that the ratio of mismatching points among all the matching points was4.4%. And the results of time statistics of the whole process of matching algorithm showed that the time for matching Red Delicious apple images of resolution ratios of0.12,0.15,0.19, and0.21mm/pixel were all less than1second in average. |
PDF Full Text Request