A multi-class classification strategy for Fisher scores: Application to signer independent sign language recognition

Abstract

Fisher kernels combine the powers of discriminative and generative classifiers by mapping the variable-length sequences to a new fixed length feature space, called the Fisher score space. The mapping is based on a single generative model and the classifier is intrinsically binary. We propose a strategy that applies a multiclass classification on each Fisher score space and combines the decisions of multiclass classifiers. We experimentally show that the Fisher scores of one class provide discriminative information for the other classes as well. We compare several multiclass classification strategies for Fisher scores generated from the Hidden Markov Models (HMMs) of sign sequences. The proposed multi-class classification strategy increases the classification accuracy in comparison with the state of the art strategies based on combining binary classifiers. To reduce the computational complexity of the Fisher score extraction and the training phases, we also propose a score space selection method and show that, similar or even higher accuracies can be obtained by using only a subset of the score spaces. Based on the proposed score space selection method, a signer adaptation technique is also presented that does not require any re-training.

Spatiotemporal-Boosted DCT Features for Head and Face Gesture Analysis

“…The first conclusion is that the proposed classifiers perform satisfactory FHG identification as they achieve scores well above 90 %. In fact, our method surpass significantly the average classification performances reported recently, i.e., 77 % in [26] and 86.4% [28] using the subset of BUHMAP dataset (210 videos of four subjects).…”

“…For recognizing these gestures, they tracked head position and rotation, then computed head velocity vector and used SVM classifiers. In Aran's study [26] a multi-class classification strategy for Fisher scores was proposed and tested on a hand gesture dataset and a sign language expression dataset [27].…”

Robust classification of face and head gestures in video

“…We can conclude that some of the proposed classifiers and especially their judiciously chosen fusions perform satisfactorily for FHG identification as they achieve scores up to 98.2%. In fact, our method surpasses significantly other publishedresults in the literature reported recently, i.e., 77% in Aran's study [33] and 67.1% in Ari's study [16] which were obtained by employing subject independent tests using only a subset of BUHMAP dataset (210 videos of four subjects).…”

“…In Aran's study [33] a multi-class classification strategy for Fisher scores was proposed and tested on a hand gesture dataset and BUHMAP [17] database. Variable length sequences were mapped to a new fixed length feature space by using a single generative model.…”

Feature selection based on cluster and variability analyses for ordinal multi-class classification problems

“…Classifiers are thus requested to maintain more powerful discriminative capability when dealing with multi-class problems. The construction of classification models for a large number of competing classes [1,4,33,43] has drawn much attention. These tasks are not handled well by general-purpose learning methods and are usually addressed in an ad hoc fashion [7,25,53].…”