Quick help: select an Image and some distance measures, and click the Go button.

Tips:

• Have more patience for the first image to be displayed.  It usually takes more time to set up the connection.
• Click the Random button to get new images randomly which could be used as the reference images in the new query.
• Select an image, and click the ZoomIn button to display that image in high resolution.
• Check two or more distance measures at one time to make your query results more precise.  You can also adjust the weights of those distance measures to make different measures having the different effect on the result.  Making a weight smaller means that the corresponding distance measure will take less important role in the query, and the number of images returned will increase in the common cases, in other word, make your query requirement loose.
• Select different method, AND, OR, or SUM to combine the checked distance measure.  AND means the images returned must satisfy all the distance measures selected.  OR means an image will be returned if it satisfy any of the distance measures selected.  SUM means the query will based on the summation of all the distance measures selected.
• When you click your mouse on an image at the first time, you will SELECT that image; the second time click means GO, or query on that image.  You can also set the meaning of the second click to ZoomIn by check the ZoomIn radio button.
• You can put an image in the Cart temporarily so that you can query on it or view it by zoomin mode later.
• You can check out the images in the cart later.  In the check out mode you can save the images into your local disk.  (only works with Netscape)

• ColorHistL14x4x4.  Based on color histogram.  Quantize the RGB cube to 4x4x4 sub-cubes and use L1 distance.
• ColorHist8x8x8.  Based on color histogram.  Quantize the RGB cube to 8x8x8 sub-cubes
• SobelEdgeHist.  Based on the histogram resulted from Sobel edge detector.
• LBPHIST.  Based on the Local Binary Partition texture measure. (texture)
• fleshiness.  Based on result from a flesh detector which can calculate the percentage of flesh in an image.
• Wavelets.  Based on the wavelet decomposition distance measure.

[comments to yi@cs.washington.edu]
Last Modified: