Identification of starch grains in microscopic images based on granulometric operations.

This study demonstrates the feasibility of feature extraction and similarity measurement for the identification of starch grains, which are often found in microscopic images of Chinese Materia Medica (CMM) and as such is an important feature for use in the authentication of CMM with abundance of starch grains and otherwise without other special microscopic characteristics. Our proposed image signature, namely, the size distribution, indicates that it is effective to characterize starch grains in microscopic images of those CMM. Experimental results in a small scale study show that the recognition rate for the proposed signature using different measures of discrepancy is around 90%, which is considered very good while the benchmark wavelet-based energy signature has only about 60% recognition. Moreover, the extraction of the size distribution only requires moderate level of computational complexity. In this study, our proposed image signature provides greater accuracy and flexibility in capturing starch grains information for the authentication of CMM, compared with other existing preliminary identification approach in which manual image segmentation is necessary.