Optimum Path Forest Approach for Image Retrieval based on Context

CBIR System consist of large datasets with millions of image samples for statistical analysis, hence putting tremendous challenge for pattern recognition techniques, which needs to be more efficient without compromising effectiveness. The image samples are stored in a database in the form of feature vectors. Pattern Recognition Technique requires a high computational burden for learning the discriminating functions that are actually responsible to separate the samples from distinct classes. Many efforts have been taken to employee machine learning algorithm in a classification problem, such as support vector machine, Artificial Neuronal Network Multi-Layer Perceptron and k-Nearest Neighbour, but all of them have usual problem of high computation burden for a training of dataset, also training becomes unrealistic due to huge training size. A novel approach is presented to reduce this problem by means of fast computation of optimum path forest in a graph derived from training samples. Each class is denoted by a multiple tree rooted at some representative samples. This Optimum Path Forest is a classifier which assigns to new sample the label of its most strongly connected root from representative samples. KeywordsOptimum Path Forest; Minimum Spanning tree; Support Vector Machine; Pattern Recognition.