Smart Internet Agent for Image Searching

•   The goal is to develop a methodology wherein a web agent can be shown a few sample images to train on.
•     Once trained, it can traverse the web to locate images of similar content.
•   A codebook is constructed at the time of agent training through clustering.  Codebooks are used then for classification.

The following images were used to build the Sky codebook: 

The Sky codebook magnified (16 codewords, 8 x 8 pixels.) 

An example of an indexed Sky image with 16% sky 

……………………………………………………………………………………

The following images were used to build the Water codebook:

The Water codebook magnified (16 codewords, 8 x 8 pixels.) 

An example of an indexed Water image with 67% water

……………………………………………………………………………………

The following images were used to build the Grass codebook:

The Grass codebook magnified (16 codewords, 8 x 8 pixels.) 

An example of an indexed Grass image with 28% grass 

…………………………………………………………………………………..

The following images were used to build the Fire codebook:

The Fire codebook magnified (16 codewords, 8 x 8 pixels.) 

An example of an indexed Red Fire image with 38% fire 

………………………………………………………………………………….

The following images used to build Light Smoke codebook:

The Light Smoke codebook magnified (16 codewords, 8 x 8 pixels.)

An example of an indexed Light Smoke image with 19% light smoke

…………………………………………………………………………………..

The following images used to build Dense Smoke codebook:

The Dense Smoke codebook magnified (16 codewords, 8 x 8 pixels.)

An example of an indexed Dense Smoke image with 11% dense smoke