Automatic Detection of In-field Defect Growth in Image Sensors

Characterization of in-field defect growth with time in digital image sensors is important for measuring the quality of sensors as they age. While more defects were found in cameras exposed to high cosmic ray radiation environments, comparing the collective growth rate of different sensor types has shown that CCD imagers develop twice as many defects as APS imagers, indicating that CCD imagers may be more sensitive to radiation. The defect growth of individual imagers can be estimated by analyzing historical image sets captured by individual cameras. This paper presents a defect tracing algorithm, which determines the presence or absence of defects by accumulating Bayesian statistics collected over a sequence of images. Recognizing the complexity of image scenes, camera settings, and local clustering of defects in color images (due to demosaicing), refinements of the algorithm have been explored and the resulting detection accuracy has increased significantly. In-field test results from 3 imagers with a total of 26 defects have shown that 96% of the defects´ dates were identified with less than 10 days difference compared to visual inspection. In addition to our continuous study of in-field defects in high-end digital SLRs, this paper presents a preliminary study of 10 cellphone cameras. Our test results address the comparison of defects types, distribution and growth found in low-end and high-end cameras with significantly different pixel sizes.