A region-disconnecting algorithm based on region growing with scanning is proposed ,which disconnects the merged rice kernel image regions through image processing. The algorithm uses the geometric feature of a single convex rice kernel to search the concave regions merged with rice kernels during the region growing with scanning and disconnect the regions at the concave positions in image. Because of using the geometric feature of rice kernels,the disconnection with the concave region identification is adaptive. In the experiment of detection of the percentage of chalky kernels, the algorithm was tested on 100 rice kernels with some physically-connected kernels and some chalky kernels,the merged rice kernel image regions were disconnected with the efficiency up to 100% ,and the detection result of the percentage of chalky kernels using the algorithm is the same as the manual inspection result.