Shikha Tripathi

Novel DCT and DWT based Watermarking Techniques for Digital Images

Two digital image watermarking techniques that have higher level of security compared to most of the existing algorithms have been proposed. The proposed digital watermarking scheme uses the properties of discrete cosine transform (DCT) and discrete wavelet transforms (DWT) to achieve almost zero visible distortion in the watermarked images. These techniques use a unique method for spreading, embedding and extracting the watermark. Embedding using a linear relation between the transform coefficients of the watermark and a security matrix has been proposed with satisfactory results. It has been shown that instead of adding the watermark to the source image, as is normally done, multiplying the watermark with the transformed source could be a preferred technique, as it would retain the original bit rate to a large extent. The watermarked images were tested subject to image cropping

Emotion Recognition from Facial Expressions Using Frequency Domain Techniques

An emotion recognition system from facial expression is used for recognizing expressions from the facial images and classifying them into one of the six basic emotions. Feature extraction and classification are the two main steps in an emotion recognition system. In this paper, two approaches viz., cropped face and whole face methods for feature extraction are implemented separately on the images taken from Cohn-Kanade (CK) and JAFFE database. Transform techniques such as Dual – Tree Complex Wavelet Transform (DT-CWT) and Gabor Wavelet Transform are considered for the formation of feature vectors along with Neural Network (NN) and K-Nearest Neighbor (KNN) as the Classifiers. These methods are combined in different possible combinations with the two aforesaid approaches and the databases to explore their efficiency. The overall average accuracy is 93% and 80% for NN and KNN respectively. The results are compared with those existing in literature and prove to be more efficient. The results suggest that cropped face approach gives better results compared to whole face approach. DT-CWT outperforms Gabor wavelet technique for both classifiers.

Real-time emotion recognition from facial images using Raspberry Pi II

In present day technology human-machine interaction is growing in demand and machine needs to understand human gestures and emotions. If a machine can identify human emotions, it can understand human behavior better, thus improving the task efficiency. Emotions can understand by text, vocal, verbal and facial expressions. Facial expressions play big role in judging emotions of a person. It is found that limited work is done in field of real time emotion recognition using facial images. In this paper, we propose a method for real time emotion recognition from facial image. In the proposed method we use three steps face detection using Haar cascade, features extraction using Active shape Model(ASM), (26 facial points extracted) and Adaboost classifier for classification of five emotions anger, disgust, happiness, neutral and surprise. The novelty of our proposed method lies in the implementation of emotion recognition at real time on Raspberry Pi II and an average accuracy of 94% is achieved at real time. The Raspberry Pi II when mounted on a mobile robot can recognize emotions dynamically in real time under social/service environments where emotion recognition plays a major role.

Systolic architecture implementation of 1D DFT and 1D DCT

Discrete Fourier Transform is widely used in signal processing for spectral analysis, filtering, image enhancement, OFDM etc. Cyclic convolution based approach is one of the techniques used for computing DFT. Using this approach an N point DFT can be computed using four pairs of [(M-1)/2]-point cyclic convolution where M is an odd number and N=4M. This work proposes an architecture for convolution based DFT and its FPGA implementation. Proposed architecture comprises of a pre-processing element, systolic array and a post processing stage. Processing element of systolic array uses a tag bit to decide on the type of operation (addition/subtraction) on the input signals. Proposed architecture is simulated for 28 point DFT using ModelSim 6.5 and synthesized using Xilinx ISE10.1 using Vertex 5 xc5vfx100t-3ff1738 FPGA as the target device and can operate at a maximum frequency of 224.9MHz. The performance analysis is carried out in terms of hardware utilization and computation time and compared with existing similar architectures. Further, as the convolution based DCT has two systolic arrays similar to that of DFT, a unified architecture is proposed for 1D DFT/1D DCT.

Dynamic facial emotion recognition from 4D video sequences

Emotions are characterized as responses to internal and external events of a person. Emotion recognition through facial expressions from videos plays a vital role in human computer interaction where the dynamic changes in face movements needs to be realized quickly. In this work, we propose a simple method, using the geometrical based approach for the recognition of six basic emotions in video sequences of BU-4DFE database. We have chosen optimum feature points out of the 83 feature points provided in the BU-4DFE database. A video expressing emotion will have frames containing neutral, onset, apex and offset of that emotion. We have dynamically identified the frame that is most expressive for an emotion (apex). The Euclidean distance between the feature points in apex and neutral frame is determined and their difference in corresponding neutral and the apex frame is calculated to form the feature vector. The feature vectors thus formed for all the emotions and subjects are given to Neural Networks (NN) and Support Vector Machine (SVM) with different kernels for classification. We have compared the accuracy obtained by NN & SVM. Our proposed method is simple, uses only two frames and yields good accuracy for BU-4DFE database. Very complex algorithms exist in literature using BU-4DFE database and our proposed simple method gives comparable results. It can be applied for real time implementation and kinesics in future.

Emotion Recognition from Facial Expressions for 4D Videos Using Geometric Approach

Emotions are important to understand human behavior. Several modalities of emotion recognition are text, speech, facial expression or gesture. Emotion recognition through facial expressions from video play a vital role in human computer interaction where the facial feature movements that convey the emotion expressed need to be recognized quickly. In this work, we propose a novel method for the recognition of six basic emotions in 4D video sequences of BU-4DFE database using geometric based approach. We have selected key facial points out of the 83 feature points provided in the BU-4DFE database. A video expressing emotion has frames containing neutral, onset, apex and offset of that emotion. We have identified the apex frame from a video sequence automatically. The Euclidean distance between the feature points in apex and neutral frame is determined and their difference in corresponding neutral and the apex frame is calculated to form the feature vector. The feature vectors thus formed for all the emotions and subjects are given to Random Forests and Support Vector Machine (SVM) for classification. We have compared the accuracy obtained by the two classifiers. Our proposed method is simple, uses only two frames and yields good accuracy for BU-4DFE database. We have determined optimum number of key facial points that could provide better recognition rate using the computed distance vectors. Our proposed method gives better results compared with literature and can be applied for real time implementation using SVM classifier and kinesics in future.

Pose invariant method for emotion recognition from 3D images

Information about the emotional state of a person can be inferred from facial expressions. Emotion recognition has become an active research area in recent years in various fields such as Human Robot Interaction (HRI), medicine, intelligent vehicle, etc., The challenges in emotion recognition from images with pose variations, motivates researchers to explore further. In this paper, we have proposed a method based on geometric features, considering images of 7 yaw angles (-45°,-30°,-15°,0°,+15°,+30°,+45°) from BU3DFE database. Most of the work that has been reported considered only positive yaw angles. In this work, we have included both positive and negative yaw angles. In the proposed method, feature extraction is carried out by concatenating distance and angle vectors between the feature points, and classification is performed using neural network. The results obtained for images with pose variations are encouraging and comparable with literature where work has been performed on pitch and yaw angles. Using our proposed method non-frontal views achieve similar accuracy when compared to frontal view thus making it pose invariant. The proposed method may be implemented for pitch and yaw angles in future.

Control of a Compass Gait Biped Robot Based on Partial Feedback Linearization

This paper deals with the control of a passivity based walking biped robot on a level surface using partial feedback linearization technique. A stable reference trajectory is generated from the Lagrangian dynamics of the biped robot and it is shown that the under actuated robot can follow the stable limit cycle trajectory using a control law based on partial feedback linearization approach. A comparison of the above approach with the conventional feedback linearization technique in terms of the Mean Square Error (MSE) is also carried out. Simulation of the forward walking is done on a compass like model of the biped robot.

Pipelined architecture for filter bank based 1-D DWT

A Convolution based parallel and pipelined architecture using MAC Loop Based Filter (MLBF) is proposed in this work. The proposed modification to the MLBF structure produces one output sample for every clock cycle as compared to the MLBF structure which produces two outputs for every four clock cycles. This results in a speed up of 2× which is significant for processing real time signals of long length. Compared to the existing MLBF based 1-D DWT architecture, proposed design uses additional 8 multipliers and 8 adders. The proposed structure is independent of the input size and filter length and performs better than other architectures with same or less area utilization. Generality, scalability, high efficiency of hardware utilization are the other merits of the proposed structure. The architecture is synthesized on Virtex 6 xc6vcx240t-2ff784 FPGA board and can operate at a maximum frequency of 633.43 MHz. The frequency of operation is twice as that of the existing approach.

Inter-Emotion Conversion using Dynamic Time Warping and Prosody Imposition

The objective of this work is to explore the importance of parameters contributing to synthesis of expression in vocal communication. The algorithm discussed in this paper uses a combination of Dynamic Time Warping (DTW) and prosody manipulation to inter-convert emotions among one another and compares with neutral to emotion conversion using objective and subjective performance indices. Existing explicit control methods are based on prosody modification using neutral speech as starting point and have not explored the possibility of conversion between inter-related emotions. Also, most of the previous work relies entirely on perception tests for evaluation of speech quality post synthesis. In this paper, the objective comparison in terms of error percentage is verified with forced choice perception test results. Both indicate the effectiveness of inter-emotion conversion by speech with better quality. The same is also depicted by synthesis results and spectrograms.