基于特征提取及聚类算法的增量图片筛选系统

Incremental image filtering system based on feature extraction and clustering algorithm

在工业生产过程中,由于需要频繁对产品进行拍照,所以会产生大量测试图片,而因为产品的外形等比较相近,因此样本图片比较相似,由此造成重复测试和测试时间过长等问题。提出一种对图片进行分类并精简数量的方法,来实现减少检测图片数量,缩短测试时间的目的。系统首先对样本图片进行特征提取,然后对特征进行降维,并对结果进行聚类来得到多个图片库,最后对图片库内的图片进行筛选,得到最终的样本库。实验对不同特征提取方法进行对比,最终从结果精度及处理时间上考虑,采取了FAST特征提取算法,有力保证了算法的质量。

In the process of industrial production, due to taking picture for the production frequently, it will make a large number of samples, and because the shape and so on are quite similar, so the sample pictures are similar, which causes the repetition test and the test time is too long. This paper presents a method to classify and reduce the number of images, to achieve the goal of reducing the number of sample images and shortening the test time. Firstly, extract the sample image feature, and then to get the feature di- mension, and take the clustering method to the pictures which are reduced dimension to get some picture libraries, in the end, sift the pictures in picture libraries to get the final sample library. Experiments on different feature extraction methods are compared, and finally from the results of the accuracy and processing time to consider, take the FAST feature extraction algorithm to guarantee the quality of the algorithm strongly.