Guiding Text Image Keypoints Extraction through Layout Analysis
Emilien Royer, Frédéric Bouchara · 2017
The rise of the smartphone industry has increased the need for mobile capture of document images and with embedded applications. In the field of image processing, detecting keypoints and computing their associated features is the first step of numerous algorithms. However, keypoints detectors are mostly designed for real world images and usually don't behave well with document images. In this paper, we propose an idea which consists in guiding keypoints extraction by using the document layout information. We compare this approach with the CORE algorithm for confusion reduction, on three classical descriptors. Results show improvements in both matching quality and processing times.