Lightweight Deep Learning Model for Defective Pixel Detection and Recovery from the Image Sensors

Abstract

Digital cameras serve as essential sensors in a wide array of applications, including self-driving cars, surveillance systems, and robotics. A critical challenge in digital imaging is the presence of defective pixels, which are small, defective dots on the sensor caused by manufacturing defects, sensor lifespan, or environmental factors. In this paper, we introduce a lightweight deep learning model designed specifically for efficient detection and recovery of defective pixels. The model comprises two main parts: the first for detecting defective pixels through simple convolutional layers and the second for recovery, utilizing a masked median filter. To reduce the complexity of the model, we employ the pixel-unshuffle and pixel-shuffle methods for feature learning on downsized features. We compare the quantitative and qualitative results of our proposed model with those of the conventional methods. The result shows that the detection result of the proposed network outperforms the previous methods.