A Mura Detection Method Based on an Improved Generative Adversarial Network

Abstract

Mura is defined as visual unevenness on the display panel. It can cause unpleasant feelings, so it is necessary to perform Mura inspection during the display quality test. However, Mura is quite difficult to be detected because of its irregular shape and size as well as its low contrast. To solve this practical problem, we proposed a GAN-based model named UADD-GAN to detect Mura in this work. Consisting of a proposed UADD generator and a discriminator, the model is trained using only normal samples, after which the generator is able to simulate the distribution of normal samples. During training, the generator takes normal images as inputs and outputs their reconstructions, while the discriminator receives images and determines whether they are original or reconstructed ones, defiantly helping the generator to perform reconstructions better. The symmetric structure and the skip-adding operation make it easy for the UADD generator to reconstruct the normal samples well. In the detecting procedure, the generator performs worse in the reconstruction of samples with Mura so that we can distinguish them from the normal ones. To make full use of discriminator, we use multiple feature layers of the discriminator for supervision instead of using only the classification layer, helping the generator to reconstruct normal samples better. Meanwhile, a two-side detecting method was used to detect Mura since all the samples are not in a square shape and it greatly improved the detecting accuracy. We have conducted experiments of Mura data sets with different proportion and our research indicates that our proposed method surpasses other state of the art methods.

Enhanced Diagnostic Accuracy in Musculoskeletal Radiography: A Comprehensive Ensemble Approach of Deep Learning Models

“…It has achieved a high accuracy of approximately 0.982. The study tells us the need for more rigorous training data and the difficulties faced in detecting slight or negligible Mura defects for further enhancements [22].…”

A Machine-Learning Strategy to Detect Mura Defects in a Low-Contrast Image by Piecewise Gamma Correction

“…It was a data augmentation method for inconspicuous targets. Xie et al [ 10 ] proposed a U-shape generator to detect Mura in a generative adversarial network (GAN), with a detection speed of ~5.6 ms. per frame and 256×256 resolution each. Moreover, the detection accuracy of defects with larger shapes was higher than that with smaller shapes.…”