Image segmentation is a fundamental process in many image, video, and computer vision applications. It is often used to partition an image into separate clusters or regions, which ideally correspond to different real-world objects. The definition of suitable similarity and homogeneity measures is a fundamental task in many important applications, ranging from geology and remote sensing to biology and medicine in the determination of the homogeneity of an organ. Image segmentation is a critical step towards visual pattern recognition and image understanding.
For years, image segmentation has been used in a supervised and an unsupervised way.1
However, unsupervised methods, which do not assume any prior scene knowledge in order to help the segmentation process, are obviously more challenging than supervised methods. In this paper, an unsupervised approach will be proposed which does not depend strongly on previously acquired information. The aim of such an unsupervised strategy is to find an appropriate segmentation process in difficult image scenes, just as is done for biomedical images.
It is known that biomedical images are often corrupted by noise and sampling artifacts, which can cause considerable difficulties when applying rigid methods. However, segmented biomedical images are now used routinely in a multitude of different applications, including diagnosis, treatment planning, localization of pathology, study of anatomic structure, and computer-integrated surgery. Thus, it is important to obtain robust segmentation methods for these types of images.
Nevertheless, in spite of the most complex algorithms developed until the present, segmentation continues to be very dependent on the application used, and there is no single method that can solve the multitude of present problems.
An example of the unsupervised segmentation method is mean shift. In this paper, we present an unsupervised strategy for biomedical image segmentation using an algorithm based on recursively applying mean shift filtering, where entropy is used as a stopping criterion.
Mean shift is a robust technique which has been used in many computer vision tasks. This is a nonparametric procedure and is an extremely versatile tool for feature analysis and can provide reliable solutions in many applications. 8
Mean shift was proposed in 1975 by Fukunaga and Hostetler,10
and largely forgotten until Cheng’s paper11
rekindled interest in it.
The term entropy is not a new concept in the field of information theory. Entropy has been used in image restoration, edge detection, and recently, as an objective evaluation method for image segmentation.12
The results obtained with the proposed algorithm are compared with manually segmented images. Our interest and main motivation for this research was to determine the robustness of our algorithm for biomedical images, while segmenting some types of lesions in an unsupervised way. In this work, lesions are the important information to be extracted from these images.