Detection, Location and Concealment of Defective Pixels in Image Sensors

International audienceThis paper presents the construction process of defective pixel detection and concealment methods, for image sensor online diagnosis and self-healing. To optimize the processing speed of the produced image, the proposed process is based on pixel neighborhood analysis using only simple arithmetic operations on the image files produced by the image sensor under test. Three detection algorithms are presented, the first one uses the distance between the pixel under test and its neighbor pixels, the second method is based on the median value of the pixel block around each pixel, and the third method is based on the ranges and the dispersion parameters in the pixel blocks of the image. The concealment process consists of substituting the defective value by the median value of the neighborhood pixel block. In the study and learning phase, distorted images obtained by injecting random disturbances into healthy reference images are used to evaluate the defective pixel detection and concealment methods. The proposed methods are compared to different state-of-the-art defective pixel detection and correction methods, in both software and FPGA implementations. The experimental results undoubtedly demonstrate that the new methods proposed in this paper perform the best results compared to the state-of-the-art