Improvement of the Input System and the Characteristics Extraction Process in Pattern Recognition of Cellular Specimens

Abstract

This paper describes the input system for pattern recognition of uterocervical cancer celles, the input system suited for the color characteristics of dyed cells, and the shading compensation based on digital technique to minimize the distortion of brightness involved in the image system. A significant density difference between background and cytoplasms, and between cytoplasms and nucleoplasms, can be achieved if an input system with color characteristics having major peaks at 600 nm and 530 nm is used. It is attempted to minimize the shading by digital technique, in place of analogue technique, which is not effective for this purpose. Compensation is carried out in individual tiny blocks, thus reducing shading to less than 4% over more than 90% of the whole screen area. For the characteristics extraction, an effective process is proposed. In conclusion, the examination method and the results of field tests using an autoscreening apparatus developed based on this process are also described.