Count cells by weighted random forest

Cell counting is always playing an important role in microscopy and many methods have been published to make the counting task automatic with high accuracy. Random forest (RF) is an effective ensemble model to implement the task. However, due to the random selection of samples and features, the base learners may differ much. The base learners with low accuracy will affect the final prediction. In this paper, a weighted averaging random forest is proposed to mitigate the problem. According to the evaluations on all training images, each base learner will be assigned a weight factor. It makes the base learners with high accuracy contribute more. The proposed weighted random forest is a linear combination of base learners with different weights and performs more robustly with accuracy on the standard VGG cells dataset.