167 APPENDIX 1 TEXTURE IMAGE DATABASES A 1.1 BRODATZ DATABASE The Brodatz's photo album is a well-known benchmark database for evaluating texture recognition algorithms. It contains 111 different texture classes and can be downloaded from http://www.ux.uis.no/~tranden/ brodatz.html. The Brodatz Album has become the de facto standard for evaluating texture algorithms, with hundreds of studies having been applied to small sets of its images. The Brodatz texture database is derived from the Brodatz Album. It was formed by cropping nine 128 x 128 subimages from the centers of 111 different original 8-bit 512 x 512 images received on tape from the Georgia Institute of Technology. Thus the database consists of 999 different 128 x 128 8-bit images, which can be considered to represent 111 different classes of data. Consequently it has a relatively large number of classes, and a small number of examples for each class. Most texture studies on classification, discrimination, and segmentation have been run on small subsets of test data from the Brodatz Album, typically four to sixteen images at once. Moreover, the tested images usually exhibit strong homogeneity within each class as well as visual and semantic dissimilarity between classes. Often they are all chosen to be microtextures. This study differs in that it includes approximately an order of magnitude greater variety, including many inhomogeneous and large-scale patterns. Additionally, the Brodatz Album has limited variety in pattern scale, rotation, contrast, and perspective. Developing
168 methods to handle these transformations is essential for recognition in real scenes, but cannot be addressed with the present Brodatz data unless it is altered. Nonetheless, the current database is significantly more diverse than has been considered in prior texture analysis studies. Consequently, it provides an important benchmark for evaluating progress in texture recognition. The first thirty texture images in the Brodatz album is shown in Figure A 1.1. Figure A 1.1 Sample Texture Images in Brodatz Album
169 A 1.2 OUTEX DATABASE The images for the experiment are downloaded from http://www.cse.oulu.fi/mvg/imagedata and used for texture classification. These texture image data is provided by Machine Vision Group (MVG) of the University of Oulu. The Machine Vision Group of the University of Oulu was established in 1981 after Matti Pietikäinen returned from his research visit to the Computer Vision Laboratory of the University of Maryland, USA. This research laboratory is one of the oldest and largest of its kind in the world. The cooperation with Maryland has continued since then, leading to many significant contributions in different areas of image analysis and computer vision. The image data for texture classification contains 32 textures from the Brodatz album. The detailed information on image dataset is given in Table A 1.1. Table A 1.1 Description of Image Dataset used for Texture Classification Type of problem Number of texture classes Number of images (samples) Image size Image format Pixel format Preprocessing Texture classification 32 textures from the Brodatz album 64 images per class (hence 2048 images in total), which comprise following subsets of images: 16 "original" images, 16 rotated versions of the "original" images, 16 scaled versions of the "original" images, 16 rotated and scaled versions of the "original" images 64x64 pixels xv (Khoros Visualization image file) 8 bit monochrome histogram equalization: each 64x64 sample has uniform gray level histogram
170 Examples of images in the dataset are shown in Figure A1.2 Figure A 1.2 Image Dataset for Texture Classification Image data for rotation invariant texture classification involving 13 textures from the Brodatz album are also provided by MVG Group and the textures are presented at 6 different rotation angles (0 o, 30 o, 60 o, 90 o, 120 o, 150 o ). Examples of images in the dataset are shown in Figure A1.3. Figure A 1.3 Image Data for Rotation Invariant Classification
171 Table A1.2. The detailed information of the image dataset is illustrated in Table A 1.2 Description of Image Dataset for Rotation Invariant Classification Type of problem Number of texture classes Rotation angles Image size Number of images (samples) Image format Pixel format Rotation-invariant texture classification 13 textures from the Brodatz album 0 o, 30 o, 60 o, 90 o, 120 o, 150 o 128x128 pixels 16 images for each class and angle (hence 1248 images in total) xv (Khoros Visualization image file) 8 bit monochrome A 1.3 FVC DATABASES Fingerprint Verification Competition (FVC) is an international competition focused on fingerprint verification software assessment. A subset of fingerprint impressions acquired with various sensors was provided to registered participants, to allow them to adjust the parameters of their algorithms. One of the most important and time-consuming tasks of any biometric system evaluation is the data collection. A multi-database was created containing four disjoint fingerprint databases, each collected with a different sensor/technology. Four distinct databases, provided by the organizers, will constitute the benchmark: DB1, DB2, DB3 and DB4. Each database is 150 fingers wide and 12 samples per finger in
172 depth (i.e., it consists of 1800 fingerprint images). Each database will be partitioned in two disjoint subsets A and B: Subsets DB1-A, DB2-A, DB3-A and DB4-A, which contain the first 140 fingers (1680 images) of DB1, DB2, DB3 and DB4, respectively, will be used for the algorithm performance evaluation. Subsets DB1-B, DB2-B, DB3-B and DB4-B, containing the last 10 fingers (120 images) of DB1, DB2, DB3 and DB4, respectively, will be made available to the participants as a development set to allow parameter tuning before the submission. During performance evaluation, fingerprints belonging to the same database will be matched against each other. The image format is BMP, 256 gray-levels, uncompressed. The image size and resolution vary depending on the database. Data collection in FVC2004 was performed without deliberately introducing difficulties such as exaggerated distortion, large amounts of rotation and displacement, wet/dry impressions, etc., The final datasets were selected from a larger database by choosing the most difficult fingers according to a quality index, to make the benchmark sufficiently difficult for a technology evaluation. The sample fingerprint images in FVC 2000, FVC 2002 and FVC 2004 datasets are shown in Figures A 1.4 to A 1.6.
173 DB1 DB2 DB3 DB4 Figure A 1.4 Sample Images from FVC2000 Figure A 1.5 Sample Images from FVC2002
174 DB1 DB2 DB3 DB4 Figure A 1.6 Sample Images from FVC2004