Human cell detection in microscopic images through discrete cosine transform and Gaussian mixture model

摘要

Automatic detection of human cell is still one of the most common investigation methods that may be used as part of a computer aided medical decision making system[1]. In this paper a statistical method based on Gaussian Mixture Model is applied to human cell detection in microscopic images[2]. 120 normal microscopic images of human cell from our research laboratory were used for analysis. Texture and grayscale features extracted from blocks of these images are given to Gaussian Mixture Model as input. It is used to model this data into three classes which are cell, extra cellular space and cell membrane [3]. Our proposed algorithm is applied on a sample dataset and experimental results show that this model is both accurate and fast with overall detection rate of around 91.23%. Error rate for cell detection was 1.82%.