Detecting dense crowds of microbes from microscope images in a global optimization framework

This paper addresses the problem of detecting a large number of densely aggregated and arbitrary oriented targets from microscope images. Scientists have long been interested in the collective behavior exhibited by aggregated organisms, and analyzing the images of hundreds of single-celled organisms may reveal the underlying biological mechanisms and biomechanics; however automatically retrieving the position and orientation of each individual in such high-density crowd is challenging and difficult for existing object detection methods. We propose in this paper a method that is able to solve this problem effectively. Contrary to conventional sliding window based detection methods that make decisions based on only local image features, the proposed method seeks to make decisions from a global perspective. Detection problem is formulated as a global optimization with all the interactions and local cues integrated in one objective function. And maximizing the objective function yields much more reasonable results. Systematical experiments have been carried out and the results demonstrate the high performance of the proposed method.