Comparative Study of Major Image Enhancement Algorithms

Abstract

Image restoration is a process of reconstruction or recovery of an image that has been corrupted or degraded by any degradation phenomenon. Image restoration techniques are inclined towards modeling the degradation and applying the inverse process in order to recover the original image. The critical goal of restoration techniques is to improve the quality of an image in some predefined manner. This present paper is a comparative study of image enhancement techniques used for improving the quality of a given image and evaluate it against the quality of a given image and evaluate it against SNR, PSNR, MSE, and SSIM as metrics.

Transfer Learning Based Traffic Light Detection and Recognition Using CNN Inception-V3 Model

“…Convolutional Neural Networks, often known as CNNs or ConvNets, are multi-stage deep architectures that combine convolutional layers with pooling or subsampling layers, followed by one or more fully connected layers. As stated by [10,21,22], its hierarchical network architecture makes it simpler to acquire invariant features and gather layer-by-layer In the illustration above, inputs are fed to develop a representation of the features using twophase convolutional and subsampling processes, and a Gaussian classifier is then used to produce a probabilistic distribution. The CNN is made up of three main components [14,23]: 1.…”

“…Each of these functions has a unique purpose. [11,10] stated that for a CNN model with binary classification, sigmoid and softmax functions are preferred, whereas softmax is typically applied for multiclass classification. According to [11,14,23], ReLU is one of the most efficient activation functions, where a nonnegative piecewise function always returns the largest value between 0 and the input, to compute, it uses the formula below.…”

“…They usually require a recognizable background and traffic light images that are at least a certain size for the algorithm to operate or assume the existence of maps that already know where every traffic light in the area is placed. Machine learning has reached a golden time of development because of the continuous development of computer technology, which has significantly reduced the limitations of artificial neural networks (ANN) [10]. In addition, significant advancements were achieved in several Iraqi Journal of Science, 2023, Vol.…”

“…Furthermore, sharing weights is used in convolution and pooling operations using sliding weighted windows. The concept of sharing weight illustrates how spatial identity, which is a statistical feature for the entire image [10], can be shared. As a result, features may be extracted from the image at each pixel point using the same weights.…”