【摘要】 文本图像二值化算法的优劣直接影响图像文本字符识别的准确率。秦简文字图像受制于背景光照欠均衡和噪声复杂等因素影响，传统文本图像二值化算法无法准确分割其前景和背景，秦简文字轮廓等特征无法准确提取，二值化效果达不到文本高准确识别要求。针对图像质量不平衡的秦简文字图像提出了一种基于图像信噪比自适应阈值模型的二值化算法。首先，将图像进行灰度转换、调整亮度和降噪等一系列二值化前的预先处理；其次，根据图像信噪比(SNR)大小自适应设置阈值，分别采用OTSU算法和Bernsen算法进行二值化处理；最后，由峰值信噪比(PSNR)与结构相似性(SSIM)评价指标择优选取二值化图像，从而准确地提取秦简图像二值化后的文字轮廓。在自建的秦简文字数据集QBS text dataset上的测试结果表明，该算法的二值化结果保留了更多的秦简文字细节特征和文字轮廓，其峰值信噪比和精确率也分别达到25.61 dB和76.67%,相较其他经典文本图像二值化算法，其性能指标均有较大提升。更多还原

【Abstract】 The quality of text image binarization algorithm directly affects the accuracy of image text character recognition. The text image of Qin bamboo slips is subject to the influence of factors such as unbalanced background illumination and complex noise. The traditional text image binarization algorithm cannot accurately segment its foreground and background, and the character contour and other features of Qin bamboo slips cannot be accurately extracted, and thus the binarization effect cannot meet the requirements of high-accuracy text recognition. In this paper, a binarization algorithm based on adaptive threshold model of image signal-to-noise ratio is proposed for Qin bamboo slips with unbalanced image quality. Firstly, the image is preprocessed before binarization, including gray-scale conversion, brightness adjustment and noise reduction; Secondly, OTSU algorithm and Bernsen algorithms are used for binarization processing, respectively, where the threshold is adaptively set according to the size of image signal-to-noise ratio(SNR); Finally, the binary image is selected according to the evaluation index of peak signal-to-noise ratio(PSNR) and structural similarity(SSIM), so as to accurately extract the text outline of the binary image of Qin bamboo slips. The test results on the self-built QBS text dataset show that the binarization result of this algorithm retains more text details and text contours of Qin bamboo slips, and its peak signal-to-noise ratio and accuracy are 25.61dB and 76.67%, respectively. Compared with other classical text image binarization algorithms, the performance indicators of the new algorithm are significantly improved.更多还原