视觉感知式场景文字检测定位方法

针对自然场景中复杂背景干扰检测的问题，本文提出一种基于视觉感知机制的场景文字检测定位方法。人类视觉感知机制通常分为快速并行预注意步骤与慢速串行注意步骤。本文方法基于人类感知机制提出一种场景文字检测定位方法，该方法首先通过两种视觉显著性方法进行预注意步骤，然后利用笔画特征以及文字相互关系实现注意步骤。本文方法在ICDAR 2013与场景汉字数据集中均取得较有竞争力的结果，实验表明可以较好地用于复杂背景的自然场景英文和汉字的检测。

Scene text detection and localization scheme with visual perception mechanism

To solve the detection problem with respect to the interference of complex backgrounds in natural scenes, in this paper, we propose a scene text detection and localization scheme based on a visual perception mechanism. The human visual perception mechanism is commonly divided into the fast parallel pre-attention step and the slow serial attention step. In our proposed scheme, we first precedes the pre-attention step with two visual saliency methods and then implement the attention step using a stroke feature and the relationship between characters. Our experimental results show the scheme to be competitive with respect to the ICDAR 2013 and the scene Chinese-character dataset. It is also suitable for English and Chinese character detection of natural scenes under complex background conditions.