Pixel Level Character Reconstruction by Background Profiling Against TMDS Emanations

In the Transition Minimized Differential Signaling (TMDS) protocol for High-Definition Multimedia Interface (HDMI), bit patterns change periodically even when the same color is displayed continuously. As a result, periodic noise appears in the reconstructed image, degrading visibility. This issue becomes particularly problematic for text composed of only a few pixels, significantly reducing the accuracy of text restoration. In response to this challenge, this study proposes a method to extract foreground text by learning the signal pattern of the background color via One-Class Classification (OCC) designed for identifying background pixels. Evaluations using an actual device confirmed that the proposed method substantially suppresses noise compared with conventional approaches, thereby improving Structural Similarity Index Measure (SSIM) and character recognition accuracy. Furthermore, under the assumption of a binary color configuration (i.e., background and foreground), readability can be maintained even for very small text.