Accord.NET Framework

This release introduces the Speeded-Up Robust Features (SURF) detector, Features from Accelerated Segment Test (FAST) corners detector, Limited-memory BFGS method for non-linear optimization, and threshold models for sequence rejection in hidden Markov sequence classifiers.

This release introduces support for independent component analysis, a new audio architecture, and a major refactoring of the hidden Markov models namespace. The new audio architecture can be used in combination with independent component analysis to perform blind source separation of audio signals. The already comprehensive set of kernels for machine learning applications has also been expanded with sparse versions of the Gaussian, Polynomial, Laplacian, Sigmoid, and Cauchy kernels.

Great improvements were made to the documentation. The framework now has support for Continuous density Hidden Markov Models, Gaussian Mixtures, and Non-negative Matrix Factorization.