Illumination Compensation and Normalization for Automatic Recognition of Rapid Diagnostic Kits

In this paper, we present an illumination compensation technique and a region of interest (ROI) extraction method for the recognition of medical diagnostic kits. The Hough transform and a multi-stage image segmentation technique are used for the ROI extraction method. To improve the accuracy of quantitative recognition, an illumination normalization method which is adaptive to the variations of background image and lighting conditions is proposed. Varying illumination condition may cause a distortion of color and intensity information of the target image. Therefore, we propose a data conversion method called illumination compensation curve which can be adjusted by a set of reference image patterns. The curve is generated by calculating the compensation factor from the change of average RGB values in the ROI of the reference image patterns. We discuss the usability of the suggested methods through the experimental results which are applied to 20 types of kit image patterns.