Accuracy Improvement of Melanosome Tracking by Error Correction

This paper proposes an error correction method for improving accuracy of melanosome tracking. Melanosomes in intracellular images are tracked manually to investigate the cause of disease, and an automatic tracking method is desirable. We detect all melanosome candidates by SIFT with 2 different parameters. Of course, the SIFT also detects non- melanosomes. Therefore, we use the 4-valued difference image (4-VDimage) to eliminate non- melanosome candidates. After tracking melanosome, we track the melanosome with low confidence again from t+1 to t. If the results from t to t+1 and from t+1 to t are different, we judge that initial tracking result is a failure, the melanosome is eliminated from candidates and re-tracking is carried out. Experimental results demonstrate that our method can correct the error and improves the accuracy.