13.   NEURALNET FILTER LONG-SEARCH
13.1   Long-Search File Structure
13.2  TradeMark  Example
13.3   Keyword-Search vs Content-Based Search


13.   NeuralNet Filter Long-Search

The limit for Short-Search is 1,000 images. There is no technical limit for the Long-Search, except your hardware limits. Long-Search can search millions of images.

The difference between Long-Search and Short-Search is the Short-Search does not look into the sub-directories and the Long-Search does look into the sub-directories.

13.1   Long-Search File Structure

The file structure looks like this:

SearchDir

 dumyfile.jpg
 Dir0000
Dir0000_000.jpg
Dir0000_001.jpg
Dir0000_002.jpg
�

Dir0000_999.jpg

Dir0001
Dir0001_000.jpg
Dir0001_001.jpg
Dir0001_002.jpg
�

Dir0001_999.jpg

Dir0002
Dir0002_000.jpg
Dir0002_001.jpg
Dir0002_002.jpg
�

Dir0002_999.jpg


In the Long-Search, the search-directory can have many sub-directories. In this version, the default number is 3,000 sub-directories. All images to be searched must be in the sub-directories. Each sub-directory can have up to 1,000 images. If your search-directory has 3,000 sub-directories and each sub-directory has 1,000 images, then you can search up to 3,000,000 images. Please leave one file in the search-directory. (In the above structure, it is �dumyfile.jpg�) The purpose of this file is for the ImageFinder to select the directory when you use the Browse button.

 All sub-directories must only be one level deep. If a sub-directory, x, has another sub-directory, y, then the images in sub-directory, y, will not be searched. If you must go more than 1 level deep, you have to use the File-Search.

When operating the ImageFinder, the only difference between Short-Search and Long-Search is to click a different command. All the parameters used are the same as the File-Search and Short-Search.

13.2  TradeMark  Example

Now we will introduce a TradeMark Search Example. The directory is �.\Uspto72\�, which has about 3,000 images.

Problem Definition:

Job:  A trademark attorney wants to make sure the trademark is new.

Problem:  The trademark in currently indexed by text. Beyond the text search, the visual search is currently very labor and time intensive.

Solution:  Attrasoft ImageFinder will exclude 99.9% of the irrelevant images in visual search, which translates into tremendous efficiency and productivity gains.
 

Figure 13.1  The key image, ./uspto72/ 72158547.gif.

Input

Key: ./uspto72/uspto_72_018/72158547.gif;
Search dir: ./uspto72/.


Parameters

NeuralNet Filter Parameters:
Blurring = 10
Sensitivity = 45
Internal Weight Cut = 30
External Cut = 300000


Operation:

Results:
Total Images: 4000
#Images Retrieved: 5
Percentage Retrieved: 0.1%

C:/�/uspto72/uspto_72_018/72158547.gif   128000000
C:/�/uspto72/uspto_72_033/72246506.gif   812500
C:/�/uspto72/uspto_72_029/72222807.gif   504000
C:/�/uspto72/uspto_72_025/72197083.gif   440000
C:/�/uspto72/uspto_72_025/72197801.gif   320000

72246506.gif   812500
 
 


 

72197083.gif   440000

Figure 13.2  Two Matched Images.

13.3   Keyword-Search vs Content-Based Search

ImageFinder provides the content-based search. The ImageFinder is further divided into:

If you need both keyword-search and content-based search, you will need to combine the Attrasoft ImageFinder with database-management-system (DBMS) software, like Microsoft Access. A DBMS application saves the meta-data, which describes the images with key words.

An imagebase or image database is an image file cabinet. An imagebase is a set of tables. A table is a set of records. A record consists of a set of fields. A field contains text or links to images.  Obviously, you can insert, update, delete, and retrieve images. The most important feature of an imagebase is to retrieve images efficiently.

You can use the DBMS software for the following functions:

You can use the ImageFinder for:
Return