Robust detection of defects in imaging arrays

As digital imagers continue to increase in size and pixel density, the detection of faults in the field becomes critical to delivering high quality output. Traditional schemes for defect detection utilize specialized hardware at the time of manufacture and are impractical for use in the field, while previously proposed software-based approaches tend to lead to quality-degrading false positive diagnoses. This paper presents an algorithm that utilizes statistical information extracted from a sequence of normally captured images to identify the location and type of defective pixels. Building on previous research, this algorithm utilizes data local to each pixel and Bayesian statistics to more accurately infer the likelihood of each defect, which successfully improves the detection time. Several defect types are considered, including pixels with one-half of the typical sensitivity and permanently stuck pixels. Monte Carlo simulations have shown that for defect densities of up to 0.5%, 50 ordinary images are sufficient to accurately identify all faults without falsely diagnosing good pixels as faulty. Testing also indicates that the algorithm can be extended to higher resolution imagers and to those with noisy stuck pixels, with only minimal cost to performance.