Improved Spatiotemporal Noise Reduction for Very Low-Light Environments

In this brief, an improved spatiotemporal noise reduction algorithm for images that are acquired through a complementary metal-oxide semiconductor image sensor (CIS) in very low-light environments is presented. To reduce the strong noises amplified by an automatic gain controller in the dark, a motion detection process is used to determine how the pixels from a spatial noise filter and a temporal noise filter are adaptively blended. The proposed algorithm also amends fixed-pattern noise, which is caused by the allowable pixel deviation of the CIS before the image is fed to the noise reduction filters. After the low-light noise is reduced, the color distortion in the low-light environments is fixed in the hue, saturation, value color space. The proposed algorithm ultimately achieved more favorable objective and subjective image qualities than the existing algorithms, as shown in the experimental results.

I. Introduction

The miniaturization of image acquisition devices has made it possible to implement video-recording devices in the design of nearly all types of smart devices, including mobile phones, portable computers, and TVs. At the center of this miniaturization is the complementary metal-oxide semiconductor (CMOS) image sensor (CIS); however, the CIS is sensitive to environmental noise signals because of a relatively small corresponding photocell. In low-light environments, both the image signals and the noise signals are mixed and are boosted by the automatic gain control (AGC), resulting in a less desirable image quality.