image-db

Matlab wrappers for various image databases.

I created this primarily as a wrapper around two natural image datasets (the van Hateren natural image dataset, and Bill Geisler's natural scene collections; links below). Each of these datasets contain multiple images, each in a separate file. Associated with each image are various items of meta data. This wrapper provides a common interface for indexing and retrieving the images and their associated meta data. Derived classes could be used to provide the same interface to any collection of files you want to index and retrieve from disk.

Example usage, assuming you have downloaded one of Bill Geisler's natural scene collections (link below):

% create the database (i.e., index the files on disk,
% parse the .exif files etc.)
db = imgdb.geisler('~/path/to/image/set1');

% select a random image/key from the dataset
key = randsample(db.keys,1);

% retrieve the image
img = db.getImg(key);

% show it...
figure; imshow(img);

% the image, img, returned above will be a scalar/grayscale
% image. This is the default behaviour for the @geisler database
% class but the images in the UT scene collections are "raw"
% colour (RGB) images. Depending on your needs, the @geisler
% database class can also retrieve the original raw image data,
% e.g.,
raw = db.getImg(key,'colorMode','raw');

% or, alternatively, can perform various colour space conversions
% on retrieval, e.g.,
xyz = db.getImg(key,'colorMode','xyz');

% or, converted to sRGB...
rgb = db.getImg(key,'colorMode','rgb');
figure; imshow(rgb);

% these conversions take into account the camera sensitiviy
% profiles, calibration parameters etc., but the images remain
% 'linear', no gamma correction is applied.

Notes

1. You can add multiple paths/directories to the database either by passing them to the constructor:

db = imgdb.geisler('~/path/to/image/set1','~/path/to/image/set2');

or explicitly calling the .add() method:

db = imgdb.geisler('~/path/to/image/set1');
db = db.add('~/path/to/image/set2');

2. There is currently nothing stopping you from adding the same path to the database multiple times. If a record already exists in the database for a given image/key (e.g., because you've already added a given path), it will be silently overwritten.

3. These wrappers use rdir() to recursively index the paths added to the database. You can get rdir() from the Mathworks File Exchange here.

Links

1. The van Hateren natural image dataset is available here.

   If you download both the linear and corrected images, the @vanhateren database will index both and provide an interface to retrieve one or the other via the optional 'corrected' argument to the .getImg() method (and/or the underlying .load() method).

2. Bill Geisler's natural image database is available as several scene collections, with extensive meta data and camera calibration notes here.

   Download a set of image files and the corresponding EXIF meta data. These are available in separate .zip files for each collection/image set on the web site. The meta data is required for the colour space conversions to work. Unzip both files in a directory of your choice and add that path to the @geisler database, either via the class constructor or using the .add() method.