The VLFeat MATLAB toolbox provides implementations of common computer vision algorithms (SIFT, MSER, AIB, KMEANS, ...) and other useful MATLAB functions, some of which are listed next:
VL_SIFT() computes the Scale-Invariant Feature Transform. SIFT bundels the most widely used feature detector and descriptor. This version is compatbile with Lowe's original implementation.
VL_MSER() computes the Maximally-Stable Extremal Regions of an image. It detects features corresponding to stable image level sets.
VL_IKMEANS() and VL_HIKMEANS() are basic implementations of K-means and hierarchical IKM optimized to work on integer data types. This is useful to quantize large collection of features, for instance to create visual dictionaries.
VL_AIB() uses the AIB algorithm to discriminatively compress a discrete random variable. An application is the compresssion of a visual dictionary for categorization.
VL_IMARRAY() and VL_IMARRAYSC() tile image stacks, VL_IMSC() scales the range of an image, VL_IMSMOOTH() convolves an image by a number of different kernels, VL_IMWBACKWARD() warps an image by backward mapping, VL_WAFFINE(), VL_WTPS(), VL_WITPS() compute various kind of image warps, VL_IMWHITEN() whitens an image, VL_XYZ2LAB(), VL_XYZ2LUV(), VL_XYZ2RGB(), VL_RGB2XYZ() perform basic color space conversions.
VL_CF() to copies a figure, VL_TIGHTSUBPLOT() provides a "borderless" version of SUBPLOT(), VL_CLICK() and VL_CLICKPOINT() interactively select image points, VL_PRINTSIZE() scales a figure printing size to a fraction of the default paper size, VL_PLOTFRAME() to plot feature frames (points, disks, oriented disks, ellipses, oriented ellpises).
VL_BINSUM() computeds binned summations (and compute histograms), VL_TWISTER() controls and runs VLFeat internal random number generator (useful to reproduce results from randomized algorithms), VL_WHISTC() computes weighed histograms, VL_GRAD() computes the gradient of a 2-D function.
See also: http://www.vlfeat.org.