An image model for quantitative image analysis

A basic assumption of most quantitative analysis techniques based on images is that each pixel within images represents a pure single class type, or object. In most cases, however, this assumption is not quite true due to finite resolutions of imaging systems. That is, pixels mapped to object boundaries may represent measurements of multiple objects. This paper presents a statistical image model which allows multiple classes within each pixel. The model assumes multichannel measurements, such as 3-channel color images, multi-spectral scanner, and thematic mapper images. Utilizing our model, we observed significant reduction in classification error and variations of quantitative measurement data