Blind bi-level image restoration with iterated quadratic programming

Many camera systems are dedicated to the capture of bi-level objects, including documents, bar codes, handwritten signatures, and vehicle license plates. Degradations in the imaging systems, however, cause blurring to the output images and introduce many more intensity levels. The blurring often arises from the optical aberrations and motions between the object and the camera, and hampers any computer vision algorithms aimed at antomatic recognition and identification of these images. While image restoration has been applied frequently in such cases, many of these algorithms do not explicitly incorporate knowledge of a bi-level object, but attempt to apply a generic restoration scheme followed by thresholding. Such two-step algorithms may not produce the best results. On the other hand, directly restoring a bi-level object is a combinatorial task and is therefore time-consuming. In this brief, we propose a method that treats the blind restoration method as an iterated quadratic programming optimization problem. This has the properties of fast convergence and good numerical stability, due to established schemes such as the interior-point algorithm. The output of our algorithm is very nearly binary. Simulation results show that by integrating the computation in the imaging system, this proposed technique can restore weak signals that would have been lost with a simple thresholding.