We propose a fast multiscale face detector that boosts a set of SVM-based hierarchy classifiers constructed with two
heterogeneous features, i.e. Multi-block Local Binary Patterns (MB-LBP) and Speeded Up Robust Features (SURF), at
different image resolutions. In this hierarchical architecture, simple and fast classifiers using efficient MB-LBP
descriptors remove large parts of the background in low and intermediate scale layers, thus only a small percentage of
background patches look similar to faces and require a more accurate but slower classifier that uses distinctive SURF
descriptor to avoid false classifications in the finest scale. By propagating only those patterns that are not classified as
background, we can quickly decrease the amount of data need to be processed. To lessen the training burden of the
hierarchy classifier, in each scale layer, a feature selection scheme using Binary Particle Swarm Optimization (BPSO)
searches the entire feature space and filters out the minimum number of discriminative features that give the highest
classification rate on a validation set, then these selected distinctive features are fed into the SVM classifier. We
compared detection performance of the proposed face detector with other state-of-the-art methods on the CMU+MIT
face dataset. Our detector achieves the best overall detection performance. The training time of our algorithm is 60 times
faster than the standard Adaboost algorithm. It takes about 70 ms for our face detector to process a 320×240 image,
which is comparable to Viola and Jones' detector.
A novel contourlet-based local feature descriptor, called Local Contourlet Binary Pattern (LCBP), is developed in this
paper. LCBP provides a multiscale and multidirectional representation for images since it integrates contourlet transform
with local binary pattern operators. Allowing for the characteristics of marginal and conditional distributions of LCBP as
well as simplicity of the model itself, we model LCBP coefficients using a two-state HMT that is in accordance with the
intra-band, inter-band and inter-direction distributions of LCBP coefficients. Based on the LCBP-HMT model, we
further propose an object recognition method that extracts parameters of the LCBP-HMT model as features and
classifies the query sample by comparing the Kullback-Liebler distance between features of the query sample and that of
the prototype objects. Experimental results illustrate the superiority of the LCBP over traditional wavelet features and
raw statistical features of contourlet coefficients in terms of the discrimination performance.
A set of new invariant moment descriptors - wavelet moment invariants, which combine wavelet multiresolution analysis and moments invariants target recognition method, are proposed in this paper and their invariance also have been proved. Wavelet moment invariants take both advantages of the wavelet inherent property of multiresolution analysis and moment invariants quality of invariant to translation, scaling changes and rotation. Furthermore, studies of the effect of using different wavelet functions and their orders are carried out. Experimental results show that wavelet moment invariants derived from the wavelet function having proper vanishing moments, symmetry and compact support have the best discrimination performance.