Image Quality and Complexity Metric for Smart CMOS Image Sensors

Abstract

Many algorithms have been developed for complementary metal oxide semiconductor (CMOS) image sensors (CIS) to speed-up analog-to-digital (A-to-D) conversion of captured images. However, there is no objective and blind image quality (IQ) metric available to compare and quantify the quality and effectiveness of these speed-up algorithms. Besides, an IQ metric is needed for intelligently control and adjust performance parameters of CIS. In this work, we present a bounded and reference-free image quality and complexity metric for these purposes. Proposed IQ metric is called conversion complexity metric (CCM). CCM is designed to quantify how complex the captured scene is, and to predict how much time and power is needed for A-to-D conversion of a captured image. The CCM concept is explained, and test results are presented for linearity, monotonicity, and percentage response to linearly changing distortion. It was found that CCM is monotonic, accomplishes 99% linearity and very responsive (316%) to distortions, providing a computationally efficient reference-free image quality metric that no other existing IQ metrics provide for CIS to intelligently adjust and optimize on-chip analog and digital signal processing operations.