Other Free Encyclopedias » Online Encyclopedia » Encyclopedia - Featured Articles » Contributed Topics from P-T

Semantic Consumer Image Indexing

images training support regions

Definition: Using Semantic Support Regions, the indexing process automatically detects the layout and applies the right tessellation template.

As a structured learning approach to represent and index consumer images with Semantic Support Regions (SSRs) (see article on Semantic Image Representation and Indexing), 26 SSRs have been designed and organized into 8 super-classes (Figure 1) from a collection of 2400 unconstrained consumer images, taken over 5 years in several countries with indoor/outdoor settings, portrait/landscape layouts, and bad quality images (faded, over-/under-exposed, blurred etc). After removing noisy marginal pixels, the images are resized to 240 × 360. The indexing process automatically detects the layout and applies the right tessellation template.

A total of 554 image regions from 138 images are cropped and 375 of them are used as training data for Support Vector Machines (SVMs) and the remaining one-third (179) for validation. Among all the kernels evaluated, those with better generalization result on the validation set are used for the indexing and retrieval tasks.Table 1 lists the training statistics of the 26 SSR classes. The negative training (test) examples for a SSR class are the union of positive training (test) examples of the other 25 classes. After learning, the SSR detectors are used to index the 2400 consumer images. Both Query By Example and Query by Spatial Icons (see article on Semantic Visual Query and Retrieval ) experiments have produced promising results .

Semantic Image Representation and Indexing -   [next] [back] Semantic Class-Based Image Indexing

User Comments

Your email address will be altered so spam harvesting bots can't read it easily.
Hide my email completely instead?

Cancel or