《视频文本检测的自动性能评价》青少年经典文学读本.pdfVIP

Automatic Performance Evaluation for Video Text Detection Xian-Sheng Hua 1, Liu Wenyin 2, Hong-Jiang Zhang 2 1 National Laboratory on Machine Perception 2 Microsoft Research China Peking University, Beijing 100871, China No. 49 Zhichun Road, Beijing 100080, China xshua@ {wyliu; hjzhang}@ Abstract interesting metrics on the output data that an algorithm or a system produces. Usually, these metrics are In this paper, we propose an objective, comprehensive expressed in terms of difference between the expected and difficulty-independent performance evaluation output and the actual output of the algorithm or system. protocol for video text detection algorithms. The They may help compare, select, improve, and even protocol includes a positive set and a negative set of design new methods to be applied in new systems indices at textbox level, which evaluate the detection designed for some specific application [9]. An overview quality in terms of both location accuracy and of PE can be found at [10]. fragmentation of the detected textboxes. In the protocol, Much research work has been done for PE in the we assign a detection difficulty (DD) level to each field of document analysis and recognition (DAR) ground truth textbox. The perfo