Abstract

Images exhibit significant variabilities that make each image different from others, even if though they belong to the same class or categories. The lack of affiliation between the heterogeneous features and the structure of images makes it challenging to model these variations for automatic image recognition and retrieval. Recognizing similar images is still a challenging task when a highly variable textured and heterogeneous features image is given as a query. In this paper, heterogeneous local penta is proposed that precisely extracts heterogeneous patterns from an image. The heterogeneity index computed over neighborhood pixels is incorporated with local penta pattern, so that it can retrieve appropriate images by extracting heterogeneous features. In this paper, a local penta pattern fused with heterogeneous features is proposed that precisely extracts heterogeneous patterns of images. The heterogeneity index computed over neighborhood pixels is incorporated with local penta pattern, so that it can retrieve appropriate images by extracting heterogeneous features. The proposed method is applied over six datasets, namely Corel 1000, Brodatz, STex, Indian movie face database, FigureQA, and our hand-crafted chart dataset. These datasets are having a highly variable and heterogeneous texture, characteristics, and properties. Experimental results depict that heterogeneous feature extracted could retrieve and recognize images with a high accuracy rate. Experimental results depict that the proposed method increases the retrieval rate by 10–15% in terms of average precision and average recall, as compared with the customary methods.

中文翻译：

---

融合异质特征的高效图像检索方法

摘要

图像显示出很大的可变性，即使它们属于相同的类别或类别，也会使每个图像与其他图像不同。异构特征和图像结构之间缺乏关联，这使得对这些变化进行建模以进行自动图像识别和检索具有挑战性。当将高度可变的纹理化和异质特征图像作为查询给出时，识别相似图像仍然是一项艰巨的任务。在本文中，提出了一种异构局部五边形，该五边形可以从图像中精确地提取出异质模式。在邻域像素上计算出的异质性指数与局部五边形模式结合在一起，因此它可以通过提取异质性特征来检索适当的图像。在本文中，提出了一种融合了异质特征的局部五边形图案，可以精确地提取图像的异质图案。在邻域像素上计算出的异质性指数与局部五边形模式结合在一起，因此它可以通过提取异质性特征来检索适当的图像。该方法被应用于六个数据集，即Corel 1000，Brodatz，STex，印度电影人脸数据库，FigureQA和我们手工制作的图表数据集。这些数据集具有高度可变且异构的纹理，特征和特性。实验结果表明，提取的异质特征能够以较高的准确率检索和识别图像。实验结果表明，该方法在平均精度和平均查全率方面将检索率提高了10-15％，