The growing popularity of neural networks encourages their use in various applications. This paper presents the classification results of distortions in the image using convolutional neural networks (CNN). This is one of the steps in various approaches to Non-Reference linage Quality Assessment, where first the distortion type is detected and then, its measured intensity is mapped to the mean opinion score. Accordingly, the score can be used as an optimization parameter for designing image processing or compression algorithms. Thus, being able to easily and reliably detect the distortion type is a very important task. The paper shows two architectures of CNN and compares the obtained results with another solution. The proposed solutions outperform other solutions, based on support vector machines, by over 10% in terms of accuracy.
Paper
Full text
Convolutional Neural Network-Based Image Distortion Classification
Semantic Scholar · Computer Science · 2019
Abstract
The growing popularity of neural networks encourages their use in various applications. This paper presents the classification results of distortions in the image using convolutional neural networks (CNN). This is one of the steps in various approaches to Non-Reference linage Quality Assessment, where first the distortion type is detected and then, its measured intensity is mapped to the mean opinion score. Accordingly, the score can be used as an optimization parameter for designing image processing or compression algorithms. Thus, being able to easily and reliably detect the distortion type is a very important task. The paper shows two architectures of CNN and compares the obtained results with another solution. The proposed solutions outperform other solutions, based on support vector machines, by over 10% in terms of accuracy.