基于图像内容的反向图片搜索引擎算法研究

发布时间：2018-08-02 13:52

【摘要】：反向图片搜索引擎是一种新型的图像搜索引擎，它可以更为直观的表达用户的搜索意图，并且能客观的帮助用户实现搜索，“以图搜图”已成为图像搜索引擎的发展方向。本文对几种典型的图像搜索算法进行了研究分析，并对各自算法的特点，以及该算法所适用的典型图片进行了深入探讨。为了适应内容丰富、信息复杂的图片搜索，满足用户对搜索结果精度的要求，对算法进行了优化。本文主要研究了基于内容的图像检索算法，包括基于颜色的图像检索算法、基于条纹的图像检索算法、基于形状的图像检索算法、基于局部特征不变的检索算法。对每种算法在实际应用中的优缺点做了分析。结合分析结果给出了基于多特征的图像检索算法。使图像检索算法的精确度和健壮性都得到了提高。主要研究内容如下：（1）改进了基于颜色特征的图像检索算法。利用HSI空间在颜色表示上的优越性结合分块提取的方法对基于颜色特征的图像检索算法做出了改善。利用分块算法在颜色的空间分布、颜色直方图表示和图像各个分块之间的联系上的优点使图像检索更加方便、准确；（2）对灰度共生矩阵、边缘直方图、Hu不变矩、角点检测等基于底层特征的图像检索算法进行性能对比，方便其在基于多特征时的应用；（3）综合各个基于底层特征的图像检索算法得出的相似度给出了加权的多特征图像检索算法和基于多特征融合的算法；（4）对实验结果进行统计分析。结果证明，基于多特征的图像检索算法较原先基于低层特征的算法，具有快速和准确等优点。更加适合于基于图像内容的反向图片搜索引擎。
[Abstract]:Reverse image search engine is a new type of image search engine, it can express users' search intention more intuitively, and can help users to realize search objectively. "search by map" has become the development direction of image search engine. In this paper, several typical image search algorithms are studied and analyzed, and the characteristics of each algorithm and the typical images suitable for the algorithm are discussed in depth. In order to adapt to the image search with rich content and complicated information, and to satisfy the user's demand for the accuracy of the search results, the algorithm is optimized. This paper mainly studies the content-based image retrieval algorithms, including color based image retrieval algorithm, striped image retrieval algorithm, shape based image retrieval algorithm and local feature invariant retrieval algorithm. The advantages and disadvantages of each algorithm in practical application are analyzed. An image retrieval algorithm based on multiple features is presented based on the analysis results. The accuracy and robustness of the image retrieval algorithm are improved. The main contents are as follows: (1) an improved image retrieval algorithm based on color features is proposed. The image retrieval algorithm based on color features is improved by using the advantages of HSI space in color representation and block extraction. Using the advantages of block algorithm in color spatial distribution, color histogram representation and image segmentation, image retrieval is more convenient and accurate. (2) for gray level co-occurrence matrix, edge histogram Hu invariant moment, Corner detection and other image retrieval algorithms based on bottom features are compared in order to facilitate their application in multi-feature based image retrieval. (3) synthesizing the similarity of the image retrieval algorithms based on the underlying features, the weighted multi-feature image retrieval algorithm and the multi-feature fusion algorithm are given. (4) the experimental results are statistically analyzed. The results show that the multi-feature based image retrieval algorithm is faster and more accurate than the previous one based on lower level features. More suitable for image content based reverse image search engine.
【学位授予单位】：长安大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.41

【参考文献】