Logo Detection Based on Convolutional Neural Networks
Chao Lu, Dandan Li, Dan Zeng · 2019 IEEE International Conference on Signal, Information and Data Processing (ICSIDP) · 2019
Logo detection processing is a popular application in images. However, due to sample variety in illumination, occlusion, rotation, and scale, the appearance of logos varies greatly, which is challenging for logo detection. In this paper, we propose a novel logo detection method based on the global pyramid text attention module. The global pyramid text attention module (GPTAM) contains the Global Text Attention Module(GTAM) and the Pyramid Text Attention Module (PTAM). GTAM can take use of the context information of texts in logos to reduce detection error, and PTAM can handle logos with multiple scales. We use non-maximum suppression(NMS) method in the postprocessing to obtain the final detection. The experiments on FlickrLogos-32 and TopLogos-10 databases demonstrate that our proposed outperforms the state-of-the-art methods. Furthermore, our method also achieves better or comparable performance on ICDAR2013 and ICDAR2015 datasets for scene text detection.