Abstract
The implementation of a previously developed method for rough image analysis for detecting watermarks in displayed video content using a smart device has been improved. The computational load involved in processing a full-size image was reduced by using Gaussian filtering to analyze the complexity of the image regions. Watermarks are assumed to be embedded in those regions with higher complexity. Regions are then checked for watermarks in descending order of complexity. Quick and reliable watermark detection is achieved by adjusting the number of regions checked so that the computational load matches device-hardware performance. Evaluation demonstrated that the improved implementation can detect watermarks at 18 fps compared with 10 fps for the previous implementation.