This paper presents an optimization-based low-light image enhancement method using spatially adaptive l₂ -norm based Retinex model. The proposed method adaptively enforces the regularization parameter using the spatially adaptive weight map, which is generated using the bright channel prior (BCP) and local variance map. Since the proposed weight map assigns the smaller weight value at the bright and edge region, the proposed method can perform weak noise reduction to preserve the edges and textures. In addition, the simplified version of the proposed method is presented using the FFT and quantized weight values for the application to consumer devices. Experimental results show that the proposed method can provide better enhanced result without the l₂ -norm minimization artifacts at the low computational cost.

A modified Hopfield neural network model for regularized image restoration is presented. The proposed network allows negative autoconnections for each neuron. A set of algorithms using the proposed neural network model is presented, with various updating modes: sequential updates; n-simultaneous updates; and partially asynchronous updates. The sequential algorithm is shown to converge to a local minimum of the energy function after a finite number of iterations. Since an algorithm which updates all n neurons simultaneously is not guaranteed to converge, a modified algorithm is presented, which is called a greedy algorithm. Although the greedy algorithm is not guaranteed to converge to a local minimum, the l (1) norm of the residual at a fixed point is bounded. A partially asynchronous algorithm is presented, which allows a neuron to have a bounded time delay to communicate with other neurons. Such an algorithm can eliminate the synchronization overhead of synchronous algorithms.

Histogram equalization is a simple and effective method for contrast enhancement as it can automatically define the intensity transformation function based on statistical characteristics of the image. However, it tends to alter the brightness of the entire image, which it is not suitable for consumer electronic products, where preservation of the original brightness is essential to avoid annoying artifacts. This paper presents a new contrast enhancement method for generalization of the existing bihistogram equalization (BHE) and recursive mean-separate histogram equalization (RMSHE) methods. The proposed method is referred to gain-controllable clipped histogram equalization (GC-CHE) to provide both histogram equalization and brightness preservation. More specifically adaptive contrast enhancement is realized by using clipped histogram equalization with controllable gain. The clipping rate is determined based on the mean brightness, and the clipping threshold is determined based on the clipping rate. The clipping rate is adaptively controlled to enhance the contrast with preserving the mean brightness. It is mathematically proven that the mean brightness of the output image converges to that of the input image with adaptive controlled. Simulation results show that the proposed GC-CHE method outperforms existing histogram-based methods, such as HE, BHE, and RMSHE, in various situations.

This paper presents a dual autoencoder network model based on the retinex theory to perform the low-light enhancement and noise reduction by combining the stacked and convolutional autoencoders. The proposed method first estimates the spatially smooth illumination component which is brighter than an input low-light image using a stacked autoencoder with a small number of hidden units. Next, we use a convolutional autoencoder which deals with two-dimensional image information to reduce the amplified noise in the brightness enhancement process. We analyzed and compared roles of the stacked and convolutional autoencoders with the constraint terms of the variational retinex model. In the experiments, we demonstrate the performance of the proposed algorithm by comparing with the state-of-the-art existing low-light and contrast enhancement methods.