Weight Optimization for Multiple Image Integration and Its Applications

Abstract

SUMMARYWe propose an image restoration technique that uses multiple image integration. The detail of the dark area when acquiring a dark scene is often deteriorated by sensor noise. Simple image integration inherently has the capability of reducing random noises, but it is especially insufficient in scenes that have a dark area. We introduce a novel image integration technique that optimizes the weights for the integration. We find the optimal weight map by solving a convex optimization problem for the weight optimization. Additionally, we apply the proposed weight optimization scheme to a single-image super-resolution problem, where we slightly modify the weight optimization problem to estimate the high-resolution image from a single low-resolution one. We use some of our experimental results to show that the weight optimization significantly improves the denoising and super-resolution performances.

Transformed-Domain Robust Multiple-Exposure Blending With Huber Loss

“…The other parameters were close to the default values shown in Table 1 of [56]. Note that we increased the number of group blocks to enhance the smoothing degree of [13], ''BM3D + HDR'': Applying BM3D [56] to the HDR image obtained by [2], ''BM3D + MEI'': the multiple-exposure images denoised by BM3D and blended by [2], and ours.…”

“…optimization method for exposure blending was proposed in [13]. In this method, an HDR image is generated by the weighted sum of a few input images with optimal weights defined for each pixel, where the weights are determined by alternately solving an optimization problem with total variation regularization [23].…”

“…Various types of noise, such as Gaussian, Poisson, impulse noise, and pixel-saturation, are contaminated in an HDR image, especially in a low-light condition. The 2 data fidelity used in previous methods, e.g., [13], is sensitive to such non-Gaussian noise, and so it cannot reduce noise sufficiently. In contrast, the use of the Huber loss function enables us to generate a high-quality HDR image in such cases.…”

“…• Weight optimization in the transform domain: We consider multiple-exposure blending in the transform (wavelet) domain to overcome the drawback of the blending in an image domain [1]- [3], [13]. The major drawback of the image domain approach is that it strictly limits the possible range of pixel values in r, e.g., if all the K pixels to be combined have negative (or positive) noise values, an ideal pixel value cannot be recovered.…”

“…This is the SNR of the HDR image tone-mapped by Reinhard et al's local operator [55]. We compared our method with the two conventional exposure-blending methods based on pixel-domain weighting; one was Debevec and Malik's method [2], i.e., an HDR image was obtained by the weighted sum with (10), and the other was the conventional weight optimization method [13]. Moreover, our results were compared with the state-of-the-art denoising method BM3D [56].…”