Preety Singh

Local Binary Pattern for automatic detection of Acute Lymphoblastic Leukemia

Acute Lymphoblastic Leukemia (ALL) is caused due to increase in number of abnormal lymphocyte cells in blood or bone marrow. This paper presents a methodology for automatic detection of the abnormal lymphocytes in a given image of the blood sample. We have used Local Binary Pattern (LBP) features for classifying the lymphocyte cell as blast or normal. LBP texture features of blood nucleus are investigated for the detection of ALL. We have also used shape features for classification and a comparative analysis of both the features is performed. It is seen that the LBP features provide reasonably good accuracy in classification.

Speaker identification using optimal lip biometrics

Biometric identification systems rely on various features for identification. Visual mouth dynamics can aid speaker recognition where the audio modality is missing or of degraded quality. This paper proposes a method of reducing the visual feature set used in a speaker-recognition system. Geometric features extracted from the lip contour are reduced using the Minimum Redundancy Maximum Relevance (MRMR) method. It is observed that MRMR features give a best accuracy of 94.7%. A small feature set reduces computation time and storage overheads. Words from a given vocabulary are also tested for speaker recognition to check robustness of MRMR features for different words.

Lipreading Using n–Gram Feature Vector

The use of n-grams is quite prevalent in the field of pattern recognition. In this paper, we use this concept to build new feature vectors from extracted parameters to be used for visual speech classification. We extract the lip contour using edge detection and connectivity analysis. The boundary is defined using six cubic curves. The visual parameters are used to build n-gram feature vectors. Two sets of classification experiments are performed with the n-gram feature vectors: using the hidden Markov model and using multiple data mining algorithms in WEKA, a tool widely used by researchers. Preliminary results show encouraging results.

NEAR-OPTIMAL GEOMETRIC FEATURE SELECTION FOR VISUAL SPEECH RECOGNITION

To improve the accuracy of visual speech recognition systems, selection of visual features is of fundamental importance. Prominent features, which are of maximum relevance for speech classification, need to be selected from a large set of extracted visual attributes. Existing methods apply feature reduction and selection techniques on image pixels constituting region-of-interest (ROI) to reduce data dimensionality. We propose application of feature selection methods on geometrical features to select the most dominant physical features. Two techniques, Minimum Redundancy Maximum Relevance (mRMR) and Correlation-based Feature Selection (CFS), have been applied on the extracted visual features. Experimental results show that recognition accuracy is not compromised when a few selected features from the complete visual feature set are used for classification, thereby reducing processing time and storage overheads considerably. Results are compared with performance of principal components obtained by application of Principal Component Analysis (PCA) on our dataset. Our set of selected features outperforms the PCA transformed data. Results show that the center and corner segments of the mouth are major contributors to visual speech recognition. Teeth pixels are shown to be a prominent visual cue. It is also seen that lip width contributes more towards visual speech recognition accuracy as compared to lip height.

Texture Features for the Detection of Acute Lymphoblastic Leukemia

Acute Lymphoblastic Leukemia (ALL) is a cancer of the blood or bone marrow. Detection of ALL is usually done by skilled pathologists, automatic detection of leukemia will reduce the diagnosis time and will also be independent of the skills of the pathologist. In this paper, we propose using texture descriptors extracted from the nucleus image for detection of ALL. The disease causes change in the chromatin distribution of the nucleus, which can be observed in the form of texture. We have used two texture features, namely Local Binary pattern and Gray Level Co-occurrence Matrix for automatic detection of ALL. A comparative analysis of both the features is presented. It is seen that LBP features perform better than GLCM features.

Lip peripheral motion for visual surveillance

Real-time surveillance systems, dealing with lipreading, can benefit from a reduction in visual data to be processed. This reduces processing time and improves the efficiency of the system. These systems take features extracted from the mouth region for recognition of speech. In this paper, the lip periphery is represented by a set of boundary descriptors. Three feature selection techniques are applied to reduce the feature set. These are Minimum Redundancy Maximum Relevance, Chi-square statistic and Correlation-based Feature Selection. Feature subsets are used for speech classification and an optimal feature vector is determined on basis of recognition performance and feature vector length. The optimal feature vector shows enhanced recognition performance while achieving a 94.17% reduction in feature size. It is observed that most of the prominent boundary descriptors lie on the upper lip. Lip width emerges as an important contributor to visual speech.

Relevant mRMR features for visual speech recognition

To improve the accuracy of visual speech recognition systems, forming a subset of relevant visual features, from a large set of extracted visual cues, is of fundamental importance. In this paper, two feature selection techniques, Principal Component Analysis (PCA) and a relatively recent method, Minimum Redundancy Maximum Relevance (mRMR), are separately applied on the extracted visual features. Prominent attributes are selected by each to form a feature vector for classification. Experimental results show that recognition accuracy for an isolated word database is not affected when a few selected mRMR features from the complete visual feature set are used for classification. This considerably reduces computation and storage overheads. It is also seen that features determined by mRMR perform better than PCA features. Both techniques yield inner mouth area segments as principal features as compared to other geometrical parameters.

Correlation based Feature Selection for Diagnosis of Acute Lymphoblastic Leukemia

Acute Lymphoblastic Leukemia (ALL) is a type of cancer characterized by increase in abnormal white blood cells in the blood or bone marrow. This paper presents a methodology to detect ALL automatically using shape features of the lymphocyte cell extracted from its image. We apply Correlation based Feature Selection technique to find a prominent set of features which can be used to predict a lymphocyte cell as normal or blast. The experiments are performed on 260 blood microscopic images of lymphocyte and an accuracy of 92.30% is obtained with a set of sixteen features.

n-Gram modeling of relevant features for lip-reading

In this paper, relevant features for a visual speech recognition system are selected using Minimum Redundancy Maximum Relevance (mRMR) method. Feature vectors, with varying number of relevant attributes, are tested to determine the most optimal feature set. It is observed that a few relevant attributes perform considerably well compared to the complete feature vector. Using this feature set as a base vector, concatenation of features is done frame-wise to build n-gram models, so as to capture the temporal behaviour of speech. It is observed that the base mRMR feature vector is able to outperform the dynamic model of n-grams. This feature vector, consisting of only significant attributes, requires less processing time due to its small size. Storage requirements are also considerably reduced.

Boundary Descriptors for Visual Speech Recognition

Lip reading has attracted considerable research interest for improved performance of automatic speech recognition (Rabiner, L., Juang, B.: Fundamentals of speech recognition. Prentice Hall, New Jersey (1993)). The key issue in visual speech recognition is the representation of the information from speech articulators as a feature vector. In this paper, we define the lips using lip contour spatial coordinates as boundary descriptors. Traditionally, Principal Component Analysis (PCA), Discrete Cosine Transform (DCT) and Discrete Fourier Transform (DFT) techniques are applied on pixels from images of the mouth. In our paper, we apply PCA on spatial points for data reduction. DCT and DFT are applied directly on the boundary descriptors to transform these spatial coordinates into the frequency domain. The new spatial and frequency domain feature vectors are used to classify the spoken word. Accuracy of 53.4% is obtained in the spatial domain and 54.3% in the frequency domain which is comparable to results reported in literature.