Recognizing cells in histological images of fish gonads

The fish fecundity in an important parameter to manage sustainable fisheries. Traditionally, its calculation is manually performed on the histological images of fish ovaries, counting and measuring its matured reproductive cells, which is a very time consuming process. The automatization of this process implies the recognition of the matured cells in the image. This paper compares the statistical performance of five state-of-art segmentation techniques, which code is publically available, to segment histological images of fish ovary in order to recognize the outline of its cells. The approaches based on Canny filter and K-means clustering provides the best trade-off in performance and speed for both fish species tested. Although other approaches provide comparable performances at pixel level, we are interested in their efficiency at region level (the recognition of the outline of the cells) in order to measure them and calculate the fish fecundity.