Abstract:In this study, we explore the potential of using a straightforward neural network inspired by the retina model to efficiently restore low-light images. The retina model imitates the neurophysiological principles and dynamics of various optical neurons. Our proposed neural network model reduces the computational overhead compared to traditional signal-processing models while achieving results similar to complex deep learning models from a subjective perceptual perspective. By directly simulating retinal neuron functionalities with neural networks, we not only avoid manual parameter optimization but also lay the groundwork for constructing artificial versions of specific neurobiological organizations.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the problem of image restoration under low - light conditions. Specifically, the authors propose a simple neural network inspired by the retina model, aiming to efficiently restore low - light images. This neural network model reduces the computational overhead compared with traditional signal - processing models, while achieving a similar effect to complex deep - learning models in subjective perception. By directly simulating the functions of retinal neurons, this model not only avoids manual parameter optimization but also lays the foundation for constructing artificial versions of specific neurobiological tissues.
### Background of the Paper
Low - light image processing has become an important research area in recent years because it plays an important role in various applications such as surveillance, self - driving vehicles, night photography, and even astronomical imaging. The challenges under low - light conditions include reduced visibility, increased noise, and loss of details, which all affect the performance of computer vision tasks (such as classification, detection, and action recognition). Although deep - learning techniques, especially convolutional neural networks (CNNs), have achieved remarkable success in low - light image processing, these models are usually large and computationally complex, limiting their practical applications in resource - constrained environments.
### Method
The method proposed in the paper is based on the working principles of different neurons in the retina, especially the functions of photoreceptors (cone cells responsible for color and rod cells responsible for brightness), horizontal cells (HCs), and amacrine cells (ACs). The model considers the optical signal - processing flow of the vertical and lateral paths in the retina and designs a simplified neural network architecture to simulate this process. This network structure maintains the independent processing of RGB channels through deep convolution and at the same time introduces a local feedback mechanism to enhance the accuracy of signal processing.
### Experiments and Results
The experiment used the publicly available LOw - Light (LOL) image dataset, which contains 500 pairs of low - light and normal - light images. The network was trained for 20 epochs with a batch size of 8 and a learning rate of 0.001. The number of network parameters is only 108, so the risk of over - fitting is small. The experimental results show that this simple network can restore the perceptual quality of low - light images to a certain extent, although it performs less well than other more complex models in objective evaluation metrics such as the Structural Similarity Index Measure (SSIM).
### Conclusion
The authors developed a simple neural network based on the functional principles of retinal neurons and applied it to the low - light image restoration (LIIR) problem. The experimental results show that, from the perspective of subjective perception, the restored images have a certain degree of satisfaction. Although it performs poorly in objective evaluation metrics, this method shows the importance of drawing inspiration from bio - computing to construct simple networks, and future work can further address the current limitations.