International Journal of Applied Pattern Recognition (9 papers in press)
A New Interval Type 2 Fuzzy based Pixel Wise Information Extraction for Face Recognition
by Sudesh Yadav, Virendra P. Vishwakarma
Abstract: A new efficient and robust approach based on pixel wise information extraction and fuzzy logic concept is presented here for the application of face recognition (FR). The motivation behind our approach is exploiting the benefit of pixel wise association of individual pixels in discriminating various classes. As all the pixels of an individual face image do not participate equally in identifying a face image from a given set of classes. Therefore, we compute the contribution made by individual pixels in identifying face images using a new interval type 2 fuzzy membership function: an extended concept of fuzzy logics. In this paper, we use a new interval type 2 fuzzy based pixel wise information extraction (NIntTy2FPIE) on input face images for computing the pixel wise association of individual pixels in recognition of a face image in a given dataset. It generates a matrix, containing value in [0, 1], where a value of a pixel in matrix near to one indicates that it is more important in discriminating the classes whereas a value near to zero indicates that it is least important. Thereafter, we apply principle component analysis (PCA) for reducing the computational cost of high dimensional data obtained above. Next, we do classification of unseen data using a variant of nearest neighbor classifier (NNC), called k-NNC. The effectiveness of NIntTy2FPIE based on k- NNC on FR systems is established with different standard datasets. Experiments performed on ORL, Yale and Georgia Tech and AR face database show that our method outperforms with many state-of-art methods and also proves that new interval type 2 membership function (MF) with k-NNC is much more efficient and robust.
Keywords: Face Recognition; Interval Type-2 Fuzzy Logics; π-Membership Function; Classification Approaches.
Artificial Neural Network Based Identification of Bharatanatyam Mudra Images using Eigen Values
by Basavaraj Anami, Venkatesh Bhandage
Abstract: This paper presents a three stage methodology for identification of Mudra images of Bharatanatyam dance. In the first stage, acquired images of Bharatanatyam Mudras are preprocessed to obtain contours of Mudras using canny edge detector. In the second stage, Eigen values are extracted as features. In the third stage, an Artificial Neural Network is used for recognition and classification of unknown Mudras. The work finds application in e-learning of Bharatanatyam dance in particular and dances in general and automation of commentary during concerts.
Keywords: contour of Mudras; Eigen value; Artificial Neural Networks; Bharatanatyam.
Human Action Recognition: A Construction of Codebook by Discriminative Features Selection Approach
by Samra Siddiqui, Muhammad Attique, Khalid Bashir, Muhammad Sharif, Faisal Azam, Muhammad Younus Javed
Abstract: Human Activity Recognition has significance in the domain of pattern recognition. HAR handles the complexity of human physical changes and heterogeneous formats of same human actions performed under dissimilar subjects. Many a novel approaches for HAR emphasize the solution for multidimensional problems. This research contributes a unique technique focusing changing human movement. The purpose is to identify and categorize human actions from video sequences. The interest points are extracted from the subject video and Motion History Images are constructed and analyzed after image segmentation. Discriminative Features are selected and the visual vocabulary is learned from the extracted discriminant features. The extracted discriminant features are then quantized by using visual vocabulary and images are represented based upon frequencies of Visual Words. The visual words are formed from the extracted discriminant features and then a histogram of visual words is developed based upon the feature vectors extracted from motion history images. These feature vectors are used for training Support Vector Machine for the classification of actions into various categories. Benchmarked datasets like KTH and Weizmann are used for evaluation and comparison with existing action recognition approaches depicts the better performance of the adopted strategy.
Keywords: Human activity recognition; foreground extraction; Feature extraction; Feature selection; Classification.
Empirical Investigation of Multiple Query Content Based Image Retrieval
by Mohamed Maher Ben Ismail, Ouiem Bchir
Abstract: Multiple query image retrieval system emerged as a promising solution to effectively understand the user interest, and communicate it to the system in order to retrieve images relevant to the user query. It consists in providing multiple example images to CBIR system in order to better reflect the information meant by the user. In the literature, multiple query based retrieval systems have been proposed. In this paper, we investigate experimentally these existing multiple query content based image retrieval systems and compare them empirically. These approaches are assessed using an image collection from Corel database. We first studied the effect of image query scoring and feature weighting.Then, we compared their performance.
Keywords: Content based image retrieval; multiple query; visual feature; comparative study.
A hybrid colour model based land cover classification using random forest and support vector machine classifiers
by M. Christy Rama, D.S. Mahendran, T.C. Raja Kumar
Abstract: Land cover monitoring using remotely sensed data requires robust classification methods for the accurate mapping of complex land cover and land use categories. Classification is a supervised learning method which maps a data item into predefined classes. Colour is an important feature used in image classification since humans tend to distinguish images mostly based on colour feature. This paper proposes a hybrid colour model for land cover classification in which colour features are extracted by combining the hue (H) values of HSV colour space and luminance (L) values of LUV colour space. The extracted features are trained and tested with random forest (RF) and support vector machine (SVM) classifiers. The performance of the proposed hybrid colour model is compared with the existing HSV colour space model using RF and SVM classifiers based on several metrics such as accuracy, sensitivity, specificity and f-score. Hyper spectral dataset of Pavia University and an IRS LISS IV orthorectified dataset are chosen as the input image for this experiment.
Keywords: colour space; classifier; decision tree; hyper spectral; orthorectified dataset.
Crowd events recognition in a video without threshold value setting
by Hocine Chebi, Dalila Acheli, Mohamed Kesraoui
Abstract: Behavioural recognition and prediction of people's activities since video present major concerns in the field of computer vision. The main objective of the proposed work is the introduction of a new algorithm which allows analysing objects in motion from the video to extract human behaviours in a complex environment. This analysis is carried out for the indoor or the outdoor environments filmed by simple means of detection (surveillance camera). The analysed scene presents in a group of people, one distinguishes the crowd scenes for an important number of people. In this type of scene, we are interested in the problems of crowd event detection by an automatic technique without setting the threshold value by neural networks to detect several anomalies in a crowd scene. To achieve these objectives, we propose a calculation of covariance and automatic artificial neural networks-based approach in order to detect several anomalies. Experiment validation has been done based on known data, where in a satisfactory results has been obtained comparing to some previous works mentioned in the state-of-the-art.
Keywords: visual analysis; crowd behaviour; intelligent video surveillance; anomaly; artificial neurons networks; ANN; automatic recognition.
Kernel-based detection of coincidentally correct test cases to improve fault localisation effectiveness
by Farid Feyzi, Saeed Parsa
Abstract: Although empirical studies have confirmed the effectiveness of spectrum-based fault localisation (SBFL) techniques, their performance may be degraded due to presence of some undesired circumstances such as the existence of coincidental correctness (CC) where one or more passing test cases exercise a faulty statement and thus causing some confusion to decide whether the underlying exercised statement is faulty or not. This article aims at improving SBFL effectiveness by mitigating the effect of CC test cases. In this regard, a new method is proposed that uses a support vector machine (SVM) with a customised kernel function. To build the kernel function, we applied a new sequence-matching algorithm that measures the similarities between passing and failing executions. We conducted some experiments to assess the proposed method. The results show that our method can effectively improve the performance of SBFL techniques.
Keywords: coincidental correctness; support vector machine; SVM; spectrum-based faults localisation; SBFL; kernel function.
A comparison of three spectral features for phone recognition in sub-optimal environments
by Sushanta Kabir Dutta, L. Joyprakash Singh
Abstract: This paper presents a comparison of three spectral features for automatic phone recognition in sub-optimal environments. An exclusive study is carried out with a phone recognition system called phonetic engine (PE) developed in the Manipuri language. The Manipuri language is a scheduled Indian language being used as the official language in the State of Manipur. However, there is no standard database of the language so far. Therefore, a PE has been built for this language. Here phonetic transcriptions are done and then modeling of each phonetic unit is carried out using hidden Markov model (HMM). Speech feature extraction is a very important stage in the development of such a PE. An analysis of phone recognition accuracies of the PE due the three dominant spectral features: MFCC, PLP and LPCC have been studied here. It is found that PLP and MFCC outperform LPCC features under all circumstances.
Keywords: mel-frequency cepstrum coefficients; MFCC; perceptual linear prediction; PLP; linear prediction cepstral coefficients; LPCC; speech features; phonetic engine; hidden Markov model; HMM; HTK toolkit.
An improved hybrid illumination normalisation and feature extraction model for face recognition
by Jyotsna Yadav, Navin Rajpal, Rajesh Mehta
Abstract: A new illumination normalisation scheme based on reflectance ratio (RR) and contrast stretching (CS), feature extraction by integer wavelet transform (IWT) in fisher subspace for face recognition is proposed in this paper. RR is ratio of pixel intensity to average pixel intensity of its neighbourhood that discards illumination effects which is followed by CS to obtain further illuminated normalised images. Robust feature extraction is acquired by selecting low frequency coefficients and ignoring high frequency coefficients using IWT which makes proposed scheme computationally efficient. Illumination normalised feature extraction by combination of RR-CS and IWT model is followed by fisher linear discriminant analysis (FLDA) to achieve robust feature vector that provides best projection direction of training and test data sets. The recognition accuracy of 100% on CMU-PIE, Yale B and 99.05% (average) on extended Yale B face databases along with comparison with state of art methods proved strength of proposed model.
Keywords: face recognition; fisher linear discriminant analysis; FLDA; illumination normalisation; integer wavelet transform; IWT; reflectance ratio; pattern recognition.