Thresholding Techniques
An important practical aim of image processing is the demarcation of objects appearing in digital images. This process is called segmentation, and a good approximation to it can be achieved by thresholding. Broadly, this process involves separating the dark and light regions of the image, and thus identifying dark objects on a light background (or the inverse). This chapter discusses the effectiveness of this idea and the means for achieving it.
• the segmentation, region-growing, and thresholding concepts.
• the problem of threshold selection.
• the limitations of global thresholding.
• problems in the form of shadows or glints (highlights).
• the possibility of modeling the image background.
• the idea of adaptive thresholding.
• the rigorous Chow and Kaneko approach.
• what can be achieved with simple local adaptive thresholding algorithms.
• more thoroughgoing variance, entropy-based, and maximum likelihood methods.
Thresholding is limited in what it can achieve, and there are severe difficulties in automatically estimating the optimum threshold—as evidenced by the many available techniques that have been devised for the purpose. Segmentation is an ill-posed problem, and it is misleading that the human eye appears to perform thresholding reliably. Nevertheless, the task can sometimes be simplified, for example, by suitable lighting schemes, so that thresholding becomes effective. Hence, it is a useful technique that needs to be included in the toolbox of available algorithms for use when appropriate. However, edge detection (Chapter 5) provides a more general means to key into complex image data.