Image Retrieval Algorithm Based Double Stage Feature Extraction and Metric

BAI Xin, WEI Lin

Packaging Engineering ›› 2018 ›› Issue (21) : 198-205.

PDF(747 KB)
PDF(747 KB)
Packaging Engineering ›› 2018 ›› Issue (21) : 198-205. DOI: 10.19554/j.cnki.1001-3563.2018.21.035

Image Retrieval Algorithm Based Double Stage Feature Extraction and Metric

  • BAI Xin1, WEI Lin2
Author information +
History +

Abstract

Aiming at the problem of information loss in semantic attribute of single low-level feature, the ability of autonomous learning is weak, which leads to poor image expression, three characteristics of Color Moment (CM), Angular Radial Transform (ART) and Edge Histogram (EH) were combined, an image retrieval scheme based on two stage feature extraction and metric was defined. Firstly, the image was transformed into HSV color space and divided into several non-overlapping sub-images. The mean, standard deviation and skewness of each sub image were calculated as CM representation. Secondly, Euclidean distance was used to extract and measure the CM of query image and database image, and output of query was marked as an image set. The ART and EH features of the query image and the image set obtained in the K results were extracted, and then the similarity between the ART and EH of the query image and the first stage image was respectively measured with Euclidean distance. By weighted combination of ART and EH, retrieval images with the highest similarity were output. Experiments showed that: compared with the current retrieval algorithms, the proposed algorithm had excellent precision of retrieval recall and superior Precision-Recall curve. This proposed algorithm has good retrieval accuracy, which has certain reference value in the fields of information processing and packaging trademark.

Cite this article

Download Citations
BAI Xin, WEI Lin. Image Retrieval Algorithm Based Double Stage Feature Extraction and Metric[J]. Packaging Engineering. 2018(21): 198-205 https://doi.org/10.19554/j.cnki.1001-3563.2018.21.035
PDF(747 KB)

Accesses

Citation

Detail

Sections
Recommended

/