Effects of Image Degradation and Degradation Removal to CNN-Based Image Classification

Abstract

Just like many other topics in computer vision, image classification has achieved significant progress recently by using deep-learning neural networks, especially the Convolutional Neural Networks (CNN). Most of the existing works are focused on classifying very clear natural images, evidenced by the widely used image databases such as Caltech-256, PASCAL VOCs and ImageNet. However, in many real applications, the acquired images may contain certain degradations that lead to various kinds of blurring, noise, and distortions. One important and interesting problem is the effect of such degradations to the performance of CNN-based image classification. More specifically, we wonder whether image-classification performance drops with each kind of degradation, whether this drop can be avoided by including degraded images into training, and whether existing computer vision algorithms that attempt to remove such degradations can help improve the image-classification performance. In this paper, we empirically study this problem for four kinds of degraded images -hazy images, underwater images, motion-blurred images and fish-eye images. For this study, we synthesize a large number of such degraded images by applying respective physical models to the clear natural images and collect a new hazy image dataset from the Internet. We expect this work can draw more interests from the community to study the classification of degraded images.

Federated Learning in Vehicular Edge Computing: A Selective Model Aggregation Approach

“…In the BU case, because the gap between the level of training image quality and the level of testing image quality is large, which leads to a low accuracy in classifying the images. The similar results appear in [5]. In the BB case, the accuracy of model aggregation with the CA approach is 2.42% higher than the accuracy of model aggregation with the FedAvg approach.…”

“…Based on the motion blur level, we try to evaluate the image quality. By [5], we consider that when the motion blur level of training images is more similar to that of testing images, the higher the classifying accuracy is resulted. As a consequence, we measure the image quality by function β that has the form as where L t is the given motion blur level of testing images.…”

“…With the rapid penetration of intelligent connected vehicles (ICV), there is an urgent need to study federated learning in VEC as an important technical framework to meet the ever-increasing demands of AI applications in vehicular networks. In this paper, we consider image classification as a typical AI application in VEC [5]. As we know, the images captured from on-board cameras usually contain sensitive information with individual privacy of the vehicular clients.…”

Visibility Restoration: A Systematic Review and Meta-Analysis

“…As mentioned previously, although there have been a few investigations on visibility restoration, none have included all essential aspects. Liu et al [4] and Pei et al [5] investigated the effects of image degradation on object recognition. The results demonstrated that the accuracy declined as haze increased.…”

Effects of objects and image quality on melanoma classification using deep neural networks

“…Some of these problems cannot be alleviated using image processing techniques. In image recognition field, there are various studies about effects of image distortions on the output [16,17,18,19,20,21]. They analyzed the strengths and weaknesses of CNN models against image degradation.…”