earticle

영어

Degradation in images is usually undesirable but unavoidable. The most common cause of this is motion. Given only the image degraded by motion blur, blind restoration aims to determine the original unblurred image and the blurring function. This paper proposes an approach to simultaneously determine these two quantities by utilising a reference point spread function (RPSF) as a learning basis in the alternating minimisation cycle. Based on the behaviour of maxima in the presence of blurs and assuming a blurring function support size, RPSF can be extracted from the given degraded image. This implies that learning is not dependent on estimated values and as a result it will not vary as restoration iteration progresses. The cost function is formulated such that spurious colour artefacts can be avoided, and the contour of the piecewise blurring function is also accounted for. Experimental results show that the proposed method is capable of determining an RPSF that closely resembles the actual blur despite the presence of noise and a small image size. Reconstruction with the proposed cost function resulted in better image quality in contrast to other methods.