An Improved Enhancement Algorithm Based on CNN Applicable for Weak Contrast Images

Abstract

Image enhancement is commonly used in digital image processing applications. Through image enhancement, the features of the target object are enhanced to make it easier to identify. In order to realize image enhancement, a neural network with a trapezoidal convolution kernel is proposed in this paper. First, weak contrast images are generated using the images in SIDD, DND and RENOIR, which are public datasets used for image denoising. On this basis, pixel distribution characteristics of the R, G and B channels of the weak contrast images are analyzed, and the histogram distribution of the images is given, which can be used to determine the size of the network convolution kernels. Then, a multi-layer convolution neural network is constructed, which uses the original image as an input and outputs the noise map. This network includes convolution kernels of different sizes for the three channels of the original images, and the convolution kernel size is determined by calculating the mean square deviation of the pixels of the corresponding channel. The proposed algorithm provides visually pleasing contrast enhancement. In this paper, a public dataset for image enhancement and actual captured images are separately assessed, both the contrast experiment and ablation experiment results show that the algorithm proposed in this paper can achieve a higher Peak Signal to Noise Ratio (PSNR) and a higher Structural Similarity Index (SSIM) than other image enhancement algorithms. INDEX TERMS Enhancement algorithm, trapezoidal convolution Kernel, image contrast, feature analysis, neural network.

Artificial Intelligence for the Metaverse: A Survey

“…In the past, many traditional image enhancement methods have been studied by applying image processing techniques, for example, histogram analysis and image decomposition [69], [70]. Recently, numerous impressive works of image enhancement have achieved considerable performance improvement by exploiting ML algorithms [71], especially DL with CNN architectures [72]- [75]. For example, a convolutional down-sampling and up-sampling network was introduced in [72] to improve the overall contrast of images, in which the deep features of RGB (red, green, and blue) channels are combined via a feature-based fusion scheme to obtain cross-channel contrast balance.…”

ICEBIN: Image Contrast Enhancement Based on Induced Norm and Local Patch Approaches

“…Some methods based on neural networks have been proposed in the field of computer vision. Wang [19] proposed a multi-layer convolutional neural network that includes convolution kernels of different sizes for the three channels of the original images. The convolution kernel size is determined by calculating the mean square deviation of the pixels of the corresponding channel.…”

Reduction of Compression Artifacts Using a Densely Cascading Image Restoration Network

“…While image compression techniques, such as joint photographic experts group (JPEG) [1], web picture [2], and high-efficiency video coding main still picture [3], can achieve significant compression performances for efficient image transmission and storage [4], they lead to undesired compression artifacts due to lossy coding because of quantization. These artifacts generally affect the performance of image restoration methods in terms of super-resolution [5][6][7][8][9][10], contrast enhancement [11][12][13][14], and edge detection [15][16][17].…”