摘要
对自然场景中的特定文字图像进行准确识别,可准确定位到所需的文字图像,提高图像检索的效率。进行特定文字图像识别过程中,需要区分文字图像和非文字图像,然后提取自然场景中特定文字图像纹理特征与边缘信息特征,传统方法不能构建文字图像识别器,难以较好地区分文字/非文字区域,降低了文字图像识别精度。提出一种深度学习的自然场景中特定文字图像优化识别模型。上述模型先融合自然场景图像的纹理特征与边缘信息特征来获得自然场景文本图像候选区,得到场景图像局部区域潜在语义识别挖掘,利用深度学习模型来表述自然场景图像中底层语义识别特征与高层语义识别之间的关系,提取不同自然场景下的语义特征,并对不同特征进行分类,利用其分类的结果组建基于深度学习的文字图像优化识别模型,利用上述模型完成对自然场景中特定文字图像优化识别。仿真结果表明,所提模型可以有效地完成自然场景中特定文字图像优化识别,具有较高的识别效率和精度。
A model of special text image optimization recognition in depth learning nature scene is proposed. The model fuses nature scene image textural feature with marginal information feature to obtain nature scene text image candidate area. It gets scene image local area potential semantic recognition excavating. Depth learning model is used to express the relation between ground floor and high level semantic recognition feature in nature scene image. It extracts semantic features in different nature scenes and classifies them. It uses the classified result to build text image optimization recognition model based on depth learning. And it uses the model to accomplish specific text image optimization recognition in nature scene. Simulation results show that above model can effectively accomplish specific text image optimization recognition in nature scene and has high recognition efficiency and accuracy.
出处
《计算机仿真》
CSCD
北大核心
2016年第11期357-360,共4页
Computer Simulation
基金
内蒙古高等学校科研项目(NJZY16383)
关键词
深度学习模型
自然场景
文字识别
Depth learning model
Natural scene
Text recognition
