Release 1.0.1
A55255-02
 Library |
 Product |
 Contents |
 Index |
| Oracle8 Visual Information Retrieval Cartridge User's Guide Release 1.0.1 A55255-02 |
|
The Visual Information Retrieval Cartridge library consists of:
The examples in this chapter assume that the stockphotos table, referred to in Chapter 3, has been created and filled with some photographic images. The table was created using the following SQL statement:
CREATE TABLE stockphotos (photo_id NUMBER, photographer (VARCHAR2(64), annotation (VARCHAR2(255), photo ORDSYS.ORDVIRB);
When you are storing or copying images, you must first create an empty BLOB in the table. Use the empty_blob( ) function, which has the following format:
ORDSYS.ORDVIRB(empty_blob(),NULL,NULL,NULL,NULL,NULL,NULL,NULL)
The Visual Information Retrieval Cartridge object data types (ODTs) are as follows:
This section presents reference information on the object data types.
For more information on BLOBs and BFILEs, see Oracle8 Application Developers Guide, Chapter 6: Large Objects (LOBs).
The ORDVirB ODT supports storage and retrieval of image data in a BLOB within an Oracle database. This object data type is defined as follows:
CREATE TYPE ORDVIRB ( -- TYPE ATTRIBUTES content BLOB, height INTEGER, width INTEGER, contentLength INTEGER, fileFormat VARCHAR2(64), contentFormat VARCHAR2(64), compressionFormat VARCHAR2(64), signature RAW(2000), -- METHOD DECLARATION MEMBER PROCEDURE copyContent(dest IN OUT ORDVIRB), MEMBER PROCEDURE setProperties(SELF IN OUT ORDVIRB), MEMBER PROCEDURE process (SELF IN OUT ORDVIRB, command IN VARCHAR2) MEMBER PROCEDURE processCopy(command IN VARCHAR2, dest IN OUT BLOB) );
Where:
In PL/SQL data is moved with the DBMS LOB package. From the client, data is moved via OCI LOB calls. The ORDVirB ODT does not supply piece-wise routines for moving data.
The ORDVirF ODT supports storage and retrieval of image data in external files (BFILEs, which are not stored in the database). BFILEs are assumed to be read-only, and this is reflected in the member procedures. This object data type is defined as follows:
CREATE TYPE ORDVIRF ( -- TYPE ATTRIBUTES content BFILE, height INTEGER, width INTEGER, contentLength INTEGER, fileFormat VARCHAR2(64), contentFormat VARCHAR2(64), compressionFormat VARCHAR2(64), signature RAW(2000), -- METHOD DECLARATION MEMBER PROCEDURE copyContent(dest IN OUT ORDVIRB), MEMBER PROCEDURE setProperties(SELF IN OUT ORDVIRF), MEMBER PROCEDURE processCopy(command IN VARCHAR2, dest IN OUT BLOB) );
Where:
This section presents reference information on the methods used for image manipulation.
The Visual Information Retrieval Cartridge methods are as follows:
For more information on object types and methods, see the Oracle8 Concepts manual.
CopyContent (dest IN OUT BLOB);
Copy an image without changing it.
The destination of the new image.
Copies the image data into the supplied BLOB.
Create a copy of the image in image1 into myblob:
image1.CopyContent(myblob);
Process (command IN VARCHAR2 );
Performs one or more image processing techniques on a BLOB, writing the image back on itself.
A list of image processing changes to make for the image.
You can change one or more of the image attributes shown in Table 4-1. See Appendix A, "File and Compression Formats" for information on the supported format combinations.
Change the file format of image1 to GIF:
image1.process('fileFormat=GIFF');
Change image1 to use lower quality JPEG compression and double the size of the image, preserving the aspect ratio:
image1.process('compressionFormat=JPEG, scale="2.0"');
Note that changing the length on only one axis (for example, xscale=2.0) does not affect the length on the other axis, and would result in image distortion.
ProcessCopy (command IN VARCHAR2,
dest IN OUT BLOB);
Copies an image BLOB to another BLOB.
A list of image processing changes to make for the image in the new copy.
The destination of the new image.
See Table 4-1, "Image Processing Parameters".
Copy an image, changing the file format, compression format, and data format in the destination image:
create or replace procedure copyit is imgB1 ORDSYS.ORDVIRB; imgB4 ORDSYS.ORDVIRF; mycommand VARCHAR2(400); begin select col2 into imgB1 from ordimgtab where col1 = 1; select col2 into imgB4 from ordimgtab where col1 = 4 for update; mycommand:= 'fileFormat=tiff compressionFormat = packbits contentFormat = 8bitlut'; imgB1.processcopy(mycommand,imgB4.content); imgB4.setproperties; update ordimgtab set col2 = imgB4 where col1 = 4; end;
SetProperties();
Writes the characteristics of an image (BLOB or BFILE) into the appropriate attribute fields.
None
After you have copied or stored an image, call this method to set the characteristics of the new image content.
This procedure sets the following information about an image:
Note that SetProperties( ) does not create the signature required for visual information retrieval. See the Analyze( ) operator in Section 4.3 for details.
Select the type, and then set the attributes using the SetProperties procedure.
imgB1 ORDSYS.ORDVIRB; . . . select col2 into imgB1 from ordimgtab where col1 = 1 for update; imgB1.setProperties; dbms_output.put_line('image width = '|| imgB1.width ); dbms_output.put_line('image height = '|| imgB1.height ); dbms_output.put_line('image size = '|| imgB1.contentLength ); dbms_output.put_line('image file type = '|| imgB1.fileFormat ); dbms_output.put_line('image type = '|| imgB1.contentFormat ); dbms_output.put_line('image compression = '|| imgB1.compressionFormat ); -- Note: signature not meaningful as displayed output.
Example output:
image width = 360 image height = 490 image size = 59650 image file type = JFIF image type = 24BITRGB image compression = JPEG
The Visual Information Retrieval Cartridge operators are located in the
ORDSYS.VIR package. The operators, which are specific to content-based retrieval, are as follows:
For ease of use, you can create a local synonym for the ORDSYS.VIR package. Connect to your schema and issue the following command:
SVRMGR> CREATE SYNONYM VIR FOR ORDSYS.VIR;
After creating the synonym, you can use it in calls to the operators: VIR.Analyze( ), VIR.Score( ), and so forth. Note that you must have the default CREATE SYNONYM privilege.
This section presents reference information on the operators.
Analyze(image IN BLOB, signature OUT RAW);
or
Analyze(image IN BFILE, signature OUT RAW);
Analyzes an image BLOB or BFILE, derives information relating to the visual attributes (including a score for each), and creates the image signature.
The BLOB or BFILE to be analyzed.
The vector to contain the signature.
The Analyze( ) operator creates the image attribute signature, which is necessary for any content-based retrieval. Whenever you are working with a new or changed image, you should also use the SetProperties( ) method to set the other image characteristics.
Signatures for facial images can be created using an optional third-party software package from Viisage Technology, Inc. After creating a facial signature, Visual Information Retrieval cartridge can convert the signature to a standard format and then compare the signatures using the Score( ) and Similar( ) operators.
Create the signatures for all images in the stockphotos table.
DECLARE temp_image ORDSYS.ORDVirB; temp_id INTEGER; cursor c1 is select id, image from stockphotos; BEGIN OPEN c1; LOOP fetch c1 into temp_id, temp_image; EXIT WHEN c1%NOTFOUND; -- Generate signature and set the properties for the image. ORDSYS.VIR.Analyze(temp_image.content, temp_image.signature); temp_image.setProperties; UPDATE stockphotos SET photo = temp_image WHERE photo_id = temp_id; END LOOP; CLOSE c1; END;
Convert(signature IN OUT RAW, operation IN VARCHAR2);
Converts the image signature to a format usable by the host machine.
The signature of the image, as created by the Analyze( ) operator. Data type is raw(2000).
The operation specifies what processing is done to the image signature. The following operations are available:
The signature is converted to the format of the host platform regardless of its initial state.
This procedure is useful if the database is stored on a remote system, but you want to do your processing locally. If your host machine is from Sun Microsystems, Inc., the Convert( ) operator sets the signature to the big endian byte order. On an Intel Corporation machine, the operator converts the signature to the little endian byte order. Note that the images themselves are platform-independent; only the signatures need to be converted.
The following example converts the signature of the image with photo_id=1 to the platform of the host system:
DECLARE myimage ORDSYS.ORDVirB; myid INTEGER; BEGIN SELECT photo INTO myimage FROM stockphotos WHERE photo_id=1; ORDSYS.VIR.Convert(myimage.signature,'BYTEORDER'); UPDATE stockphotos SET photo=myimage; END;
Score(signature1 IN RAW,
signature2 IN RAW,
weightstring IN VARCHAR2);
where weightstring is: 'globalcolor="val", localcolor="val", texture="val", structure="val"'
or: 'facial=1'
Compares the signatures of two images and returns a number representing the weighted sum of the distances for the visual attributes.
The signature of the comparison image (the image with which other images are being compared to test for matches). Data type is raw(2000).
The signature of the image being compared with the comparison image. Data type is raw(2000).
A list of weights to apply to each visual attribute. Data type is VARCHAR2. The following attributes can be specified:
This function returns a FLOAT value between 0.0 and 100.0.
Before the Score( ) operator can be used, the image signatures must be created with the Analyze( ) operator.
The Score( ) operator can be useful when an application wants to make finer distinctions about matching than the simple "Yes" or "No" returned by Similar( ). For example, using the number for weighted sum returned by Score( ), the application might assign each image being compared to one of several categories, such as Definite Matches, Probable Matches, Possible Matches, and Non-Matches. The Score( ) operator can also be useful if the application needs to perform special processing based on the degree of similarity between images.
The weights supplied for the four visual attributes are normalized prior to processing such that they add up to 100 percent. To avoid confusion and meaningless results, you should develop a habit of always using the same scale, whether 0 to 100 or 0.0 to 1.0.
You must specify at least one of the four image attributes or the facial attribute, in the weightstring. You cannot combine the facial attribute with any of the other attributes.
The following example finds the weighted sum of the distances between image 1 and the other images in the stockphotos table, using the following weights for the visual attributes:
This example assumes that the signatures have already been created using the Analyze( ) operator and they are stored in the database.
DECLARE weighted_sum NUMBER; BEGIN SELECT Q.photo_id, ORDSYS.VIR.Score(S.photo.signature, Q.photo.signature, 'globalcolor="0.2"
localcolor="0.2"
texture="0.1"
structure="0.5"') weighted_sum FROM stockphotos Q, stockphotos S WHERE S.photo_id=1 and Q.photo_id !=S.photo_id; END;
The following shows possible output from this example. The last image has the highest score, and therefore is the best match of the test image (photo_id=1). Changing the weights used in the scoring would lead to different results.
PHOTO_ID WEIGHTED_SUM ---------- ------------- 2 3.79988 3 76.0807 4 47.8139 5 80.451 6 91.2473 5 rows selected.
Similar(signature1 IN RAW
signature2 IN RAW,
weightstring IN VARCHAR2
threshold IN FLOAT);
where weightstring is: 'globalcolor="val", localcolor="val", texture="val", structure="val"'
or: 'facial=1'
Determines whether or not two images match. Specifically, compares the signatures of two images, computes a weighted sum of the distance between the two images using weight values for the visual attributes, compares the weighted sum with the threshold value, and returns the integer value 1 if the weighted sum is less than or equal to the threshold value, and otherwise returns 0.
The signature of the comparison image (the image with which other images are being compared to test for matches).
The signature of the image being compared with the comparison image.
A list of weights to apply to each visual attribute. Data type is VARCHAR2. The following attributes can be specified:
The threshold value with which the weighted sum of the distances is to be compared. If the weighted sum is less than or equal to the threshold value, the images are considered to match. This range of this parameter is from 0.0 to 100.0.
This operator returns an integer value of 0 (not similar) or 1 (match).
Before the Similar( ) operator can be used, the image signatures must be created with the Analyze( ) operator.
The Similar( ) operator is useful when the application needs a simple "Yes" or "No" for whether or not two images match. The Score( ) operator is useful when an application wants to make finer distinctions about matching or to perform special processing based on the degree of similarity between images.
The weights supplied for the four visual attributes are normalized prior to processing such that they add up to 100 percent. To avoid confusion and meaningless results, you should develop a habit of always using the same scale, whether 0 to 100 or 0.0 to 1.0.
You must specify at least one of the four image attributes, or the facial attribute in the weightstring. You cannot combine the facial attribute with any of the other attributes.
This example checks the first image against all of the other images in the table and determines if there are any matches, using a threshold value of 75 and the following weights for the visual attributes:
DECLARE BEGIN SELECT Q.photo_id FROM stockphotos Q, stockphotos S WHERE S.photo_id=1 and Q.photo_id != S.photo_id AND ORDSYS.VIR.Similar(S.photo.signature, Q.photo.signature, 'globalcolor="20"
localcolor="20"
texture="10"
structure="50"', 75)=1; END;
The following shows a possible output from this example. See the Example section of the Score( ) operator for a different result using the same images and weights.
PHOTO_ID ---------- 3 5 6 3 rows selected.
