TV screen content extraction and recognition algorithm for the verification of digital television systems

Abstract

This paper presents an algorithm for detecting the TV screen content and comparing it with the referent contents. The TV screen content is extracted in two steps. The first step is the TV screen edge detection, performed with Scharr edge detection, detection of long lines and TV screen rectangle detection. The second step is the extraction of the TV screen content between the detected edges and scaling it to the dimensions of the referent image. The paper presents two methods for comparison of the detected content with the referent images: least-absolute-error method and normalized cross-correlation method. The methods were tested under constant and variable illumination conditions. The least-absolute-error method was successful under both conditions, while the normalized cross-correlation showed vulnerability for small differences under variable illumination. The algorithm can be used in test systems for digital television systems. After performance improvements it can also be used in consumer electronic industry. © 2010 IEEE.