MATLAB CVIP Toolbox Functions

Pattern Classification: Normalization Methods

The pattern classification functions provide a valuable tool for the development of computer vision algorithms. It is a large group with separate function categories for feature extraction, distance and similarity metrics, data normalization, and the classification algorithms. In general there are two types of functions – those that operate on individual objects or individual feature vectors and those that operate on groups of images, objects or feature vectors. The higher level functions that operate on groups use matrices, or comma separated value (csv) files, to pass information between functions. The format of the csv files is similar to the CVIPtools feature files (for details see the Help documentation).

The feature functions available include binary object (shape) features, histogram features, texture features and spectral features. City block, Euclidean, maximum value, and Minkowski distance metrics as well as Tanimoto and vector inner product similarity metrics are available for comparison of feature vectors. For data normalization we include min-max, standard normal density (SND), range based, softmax scaling and unit vector normalization, or the user can opt to perform no data normalization. The pattern classification methods we include are the basic methods: nearest neighbor, K nearest neighbor and nearest centroid. Note that MATLAB has a variety of related functions of interest that can be used in conjunction with our functions. Neural networks, support vector machines, genetic learning algorithms and many others are available for pattern classification, depending on the other toolboxes that are installed.

min_max_norm_cvip - applies min-max normalization to set of feature vectors in a matrix
pattern_min_max_norm_cvip - two csv files are input, training and test sets, and returns new csv files with the feature vectors normalized with min-max normalization
pattern_range_norm_cvip - two csv files are input, training and test sets, and returns new csv files with the feature vectors normalized with range normalization
pattern_snd_norm_cvip - two csv files are input, training and test sets, and returns new csv files with the feature vectors normalized with standard normal density (snd) normalization
pattern_softmax_norm_cvip - two csv files are input, training and test sets, and returns new csv files with the feature vectors normalized with softmax scaling normalization
pattern_unit_vector_norm_cvip - two csv files are input, training and test sets, and returns new csv files with the feature vectors normalized with unit vector normalization
range_norm_cvip - normalizes a set of feature vector values in a matrix based on the range of each feature
softmax_cvip - applies softmax normalization to set of feature vectors in a matrix
snd_norm_cvip - applies standard normal density normalization to a set of feature vectors in a matrix
unit_vector_norm_cvip - applies unit vector normalization to set of feature vectors in a matrix