Relevance Feedback Approach For Trademark Image Retrieval Using Query Improvement Strategy

Volume 13 Issue 2 November - January 2018

Research Paper

Relevance Feedback Approach For Trademark Image Retrieval Using Query Improvement Strategy

Latika Pinjarkar*, Manisha Sharma**, Smita Selot***

* Associate Professor, Department of Information Technology, Shri Shankaracharya Technical Campus, Bhilai, India.

** Professor and Head, Department of Electronics and Telecommunication, Bhilai Institute of Technology, Durg, Chhattisgarh, India.

*** Professor and Head, Department of Computer Applications, Shri Shankaracharya, Technical Campus, Bhilai, India.

Pinjarkar, L., Sharma, M., and Selot, S. (2018). Relevance Feedback Approach For Trademark Image Retrieval Using Query Improvement Strategy. i-manager’s Journal on Future Engineering and Technology, 13(2), 17-27.https://doi.org/10.26634/jfet.13.2.13868

Abstract

This paper proposes an automated system for rotation, scaling, and translation invariant trademark retrieval based on colored trademark images. Trademark images are recognized using color, shape, and texture feature extraction. Color Feature extraction is done by implementing Color Histogram, Color Moments, and Color Correlogram techniques. Texture features are extracted by Gabor Wavelet and Haar Wavelet implementation. Shape feature extraction is implemented by using Fourier Descriptor, Circularity features. The proposed trademark retrieval approach uses Relevance Feedback and three kinds of query improvement strategies, New Query Point (NQP), Query Rewriting (QRW), and Query Development (QDE). The datasets used for experimentation are publicly available FlickrLogos 27 and FlickrLogos 32 databases. The query image is varied from the database images, by applying transformations on the query image like rotation, scaling, and translation of the image by number of pixels in X and Y direction. The system is tested for transformed query images with different combination of transformations. The proposed system is highly robust giving good retrieval results against these transformations.