一种基于混合区域分割的特征词袋模型识别算法

数字广告图像识别与数据分析在广告精准推送、收视行为分析、商业网络搜索等各大应用领域有着重要的应用。针对网络图片搜索的实际需求，结合图像特征提取识别技术，本文对数字媒体广告图像的识别和分类方法进行了研究。针对传统的基于空间金字塔匹配的视觉词典模型算法匹配条件较为苛刻、缺乏抗形变能力等局限性，在视觉词典模型算法的基础上，引入视觉特征词典特征点的空间位置信息，使用混合高斯模型实现了自适应的空间区域划分。实验结果表明，识别算法的抗形变能力和识别精度在标准库及本地广告库的测试中均能取得较好的测试结果。

A Word Bag Recognition Algorithm Based on Mixture Region Segmentation

Digital advertising image recognition and data analysis have a wide application in precise advertising push, viewing behavior analysis, and business web search. This paper studys the recognition and classification of digital media advertising images by analyzing the actual characteristics of advertising images and utilizing image feature extraction technology. Aiming at the shortcomings of the traditional visual word bag based on spatial Pyramid matching algorithm, e.g., the harsh matching condition and the lack of deformation resistance, this paper uses Gaussian mixture model to achieve the adaptive spatial zoning, by introducing the spatial location information of the feature points of the visual word bag. It is shown that the deformation resistance ability and the recognition accuracy in the proposed algorithm can attain good results in the test of the standard database and the local advertising database.