改进的Gabor小波变换特征提取方法

摘要

针对自然场景中文本存在图像背景复杂、颜色多变、噪声强烈,图像存在变形、残缺、模糊、断裂等问题,提出一种基于方向预分类的Gabor小波变换特征提取方法.利用Gabor函数良好的频率选择性和方向选择性,同时考虑到笔划相对位置的偏移,方向预分类使得滤波器对笔划方向的选择更有针对性.实验结果证明,该特征提取方法对笔划变形和低分辨率字符具有较好的适应性,能有效解决低质量场景文本的识别问题.%Gabor wavelet transformation feature extraction method based on direction pre-classification is presented to resolve the recognition of scene text which is together with complex background, inconsistent of color, strong noise, deformation, incomplete, blur, fracture and so on. The method uses the frequency and direction selectivity of Gabor function, and fully considers the direction offset. Direction pre-classificatiou makes the stroke direction selection more accurate. Experimental result shows the method is adaptive for character of stroke deformation or low resolution and can recognize low-quality scene text effectively.