Stitching Based on Corrections to Obtain a Flat Image on a Curved-Edge OLED Display

Vision-based inspection, measurement, and repair technologies are becoming extremely significant in the manufacturing of display screens. These allow for concurrent luminance measurement and defect detection across millions of display subpixels. However, for curved organic light-emitting diode (OLED) displays, the image quality of curved edges is too poor and the localization of subpixel-level pixel or defects cannot be achieved. As a result, detection and measurement algorithms or systems that would otherwise work on flat screens fail when rolled out to curved screens. In this article, we proposed a Stitching-Based-on-Corrections (SBoC) method to obtain a standardized flat image on OLED displays with curved edges. This allows traditional methods to become applicable again. First, reflective prisms were added to the imaging system to capture clear images of the curved-edge regions of the display screen. Then, an active calibration image flattening algorithm based on polynomial geometric correction was proposed to standardize the images on the display screen. Second, we designed adaptive gray-level corrections perpendicular and parallel to the principal axis of the reflective prisms to restore the brightness and contrast of curved-edge regions to an ideal flat state. The corrected images of different regions were then stitched together, and an ideal, distortion-free image was produced, with pixels distributed in a strict regular pattern. The experimental results confirmed the effectiveness of the proposed method.

I. Introduction

Vision-based inspection [1], [2], measurement [3], and repair [4], [5] technologies are central to modern-day display screen manufacturing. These technologies permit large-scale, concurrent luminance measurement [6] and defect detection [7] for millions of display subpixels. However, for curved organic light-emitting diode (OLED) displays, which are increasingly in demand in the market, the image quality of the curved edges is too poor and exhibits fuzziness, distortion, and brightness decay. Consequently, subpixel-level pixel or defect localization cannot be achieved.