To search, Click below search items. |
| All Published Papers Search Service |
|
|
|
Title |
Domain Specific Content Based Image Retrieval (CBIR) for Feminine Textile Designs |
|
Author |
Humera Tariq, Saad Sheikh, Mohsin Ali, Usman Amjad, Ahsan Ali |
| Citation |
Vol. 19 No. 2 pp. 88-101 |
|
Abstract |
Parsing of color and texture into machine understandable pattern is an important step for bringing in satisfactory results in response to a particular query image. The objective of the paper is to investigate the problem of storing, indexing and retrieval of challenging eastern feminine fashion shoots on the basis of low level visual descriptors. We introduce a novel domain specific dataset of 1500 challenging images with a large variation in pose and background from fashion and textile industry the images are heavily textured with enormous color variations. Human detection is performed using HOG and Hard Negative Mining on fashion photographs. Training has been performed through Multiscale and Multiclass SVM to obtain LBP features for texture classification. Re-allocation is used to improve texture classification. True Positive Rate (TPR) for human detection with skeletal images, mining on positive images and mining with negative images are found to be 65%, 21% and 55% respectively Color retrieval accuracy is greater than 90% in multidimensional HSV space. Qualitative Results for texture classification are up to the mark for gray color design retrieval. |
|
Keywords |
Visual descriptor, Hard Negative Mining of Oriented Gradients (HOG), Support Vector Machine (SVM), Local Binary Pattern (LBP), Textile, Fashion Industry, Texture classification. |
|
URL |
