Current Issue Cover


摘 要
目的 文档图像检索过程中,传统的OCR技术因易受文档图像质量和字体等相关因素的影响,难以到达有效检索的目的。关键词识别技术作为OCR技术的替代方案,不需经过繁琐的OCR识别,可直接对关键词进行检索,并取得了不错的结果。而且在关键词提取过程中,局部特征提取相对于全局特征提取而言,有着更详细、更准确的描述,在角点检测方面,本文针对Harris算法聚簇现象严重和运算速度慢等问题。方法 在关键词识别技术的框架下提出了改进Harris的图像匹配算法,该算法是第一次运用在文档图像检索中,首先基于Fast进行特征点的检测,很大程度上减少了在特征检测上耗费的时间复杂度,然后利用Harris进行特征描述,采用非极大值抑制的方法,不仅提高了原始Harris算法的实时性,还明显改善了聚簇现象严重的问题,最后利用暴力匹配中的汉明距离进行特征的相似性度量,输出最终的匹配结果。结果 实验结果表明在无噪声的情况下,准确率达到98%,召回率达到87.5%,而且在固定均值,不断提高方差的高斯噪声条件下,也取得了较好的实验效果。与Sift算法相比,时间也得到了明显的提高,准确率也高于后者。结论 本文提出的方法满足了快速、精确的查找需求,在印刷体图像的文档图像检索中有效提高了检索率,具有较好的实验效果。
Printed image retrieval based on improved harris feature

Gao Ting,Askar Hamdulla,Abdusalam Dawut(Institute of Information Science and Engineering, Xinjiang University)

Objective Today, in the 21st century, with the rapid development of Internet information, while providing great convenience for people"s lifestyles, people also have to face the redundancy of information when they go online, because most of the information is now in text. Forms exist, so how to accurately and efficiently obtain the information that users need is especially important. Moreover, with the acceleration of the informationization process, the number of electronic documents has also risen sharply, so it is becoming more and more urgent to perform efficient and fast retrieval of document images at this time. In the process of document image retrieval, traditional OCR technology is difficult to achieve effective retrieval due to the quality of document images and fonts. As an alternative to OCR technology, word recognition technology does not require cumbersome OCR recognition, and can directly search for keywords to achieve good results. In addition, in the keyword extraction process, local feature extraction has a more detailed and accurate description than global feature extraction. In terms of corner detection, this paper focuses on the problem of serious clustering and slow computing speed of Harris algorithm. Method In the framework of Word Spotting technology, an improved Harris image matching algorithm is proposed, which is used for document image retrieval for the first time. Firstly, the original Harris algorithm uses the Gaussian function to smooth the window in the feature point extraction process of the image. When calculating the corner response value R, the differential operator is used as the directional derivative, so that the number of multiplication operations will be calculated in the calculation process. More, resulting in a large amount of computational algorithms, slow operation and other issues. In view of the deficiencies of the original Harris algorithm, the Fast algorithm is used in the detection of corner points. (1) The gray value of the center pixel and the surrounding 16 pixels is compared according to the formula in the radius 3 field. (2) In the detection process, in order to improve the detection speed, we generally first detect the 0th and 8th pixel points on the circumference, and then sequentially detect the two points on the other diameters.(3) If there is a difference between the gray value of 12 consecutive points and the gray value of the p point exceeding the threshold, it is a corner point.(4) After obtaining the primary corner, the Harris algorithm is used to remove the pseudo corner. Secondly, the original Harris algorithm sorts and compares the local maximum of the corner response function, then establishes the response matrix and the coordinate matrix, records the local maximum and the response coordinates respectively, and compares the global maximum. At this point, all the corner points have been recorded, but it is very likely that there will be a case where multiple corner points coexist in the domain of a corner point, that is, "clustering" phenomenon. Aiming at the serious problem of clustering of Harris algorithm, a non-maximum value suppression method is adopted, which is essentially to search for local maximum and suppress non-maximum elements. When detecting the diagonal points, the local maximum is sorted from large to small and the suppression radius is set, a new response function matrix is established, and the corner points are extracted by continuously reducing the radius, thereby effectively avoiding the phenomenon of Harris corner clustering. (1) Calculate the value of the corner response function of all the pixels in the graph, and search for the local maximum, and record the pixel of the local maximum. (2) Establish a local maximum ordering matrix and corresponding coordinate matrix, and sort the local maximums from large to small.(3) Set the suppression radius to the local maximum, establish a new matrix of response functions, and extract the corner points by continuously reducing the radius.(4) It is judged whether the local maximum value is the maximum value within the suppression radius, that is, the desired corner point, and if the condition is satisfied, the local maximum value is added to the response function matrix.(5) Continue to reduce the radius to extract the corner points. If the number of corner points is preset, the process ends. Otherwise, repeat step (4). Result The experimental results show that the accuracy rate is 98% and the recall rate is 87.5% without noise, and good experimental results are obtained under the condition of constant mean and continuous improvement of variance of Gaussian noise. Compared with the Sift algorithm, the time has also been significantly improved, and the accuracy is also higher than the latter. Conclusion Starting from the document image of printed matter, using Fast+Harris to search for keywords under the framework of keyword recognition technology, on the one hand, it saves the retrieval time, improves the real-time performance of the algorithm, and on the other hand improves the aggregation of Harris. The cluster phenomenon improves the accuracy of corner detection. Compared with the Sift algorithm, not only the time is greatly improved, but also under the influence of different degrees of noise, it has achieved good experimental results.