Research

Human health monitoring employing facial and corporeal-based visual feature

Face and corporeal-based based visual features are highly rich in reflecting the cognitive and physical condition of an individual. Therefore, these visual features can be used to automate diagnoses and additionally can simplify the diagnosis for a certain scenario where these examinations can time-consuming involving medical personnel. One such scenario is examining severely demented Alzheimer’s disease (AD) patients. It is essential to diagnose their empathy and expressiveness since such patients have lost a substantial amount of their cognitive capacity along with certain symptoms like apathy. Diagnosing such symptom at an early stage can be very effective for treatment.

Thus large-scale monitoring these scenarios for AD patients is costly and logistically inconvenient for patients and clinical staff. Therefore in these research approaches need to cater to the challenging settings of current medical recordings, which include continuous pose variations, occlusions, camera-movements, camera-artefacts, as well as changing illumination. Additionally and importantly, the (elderly) patients exhibit generally less profound facial activities and expressions in a range of intensities. Hence we employed the various key component and models for emotion-based on the computer vision to gauge their expression feature, followed by statistical sampling to process the expression-based feature, and finally deploying machine learning based classification model to obtain the final diagnosis.

Related project: Inria and CAS project on Facial Expression Recognition with application in human health monitoring (FER4HM)

Adaptability and liveness in multimodal ocular biometrics

Multimodal ocular biometrics using sclera and iris has not been extensively studied. Therefore, the first phase of the research concentrates on designing an image processing and pattern recognition module for evaluating the potential of the sclera and its combination with the iris biometric. To investigate this, variously advanced image segmentation, enhancement, feature extraction, classification and fusion techniques are employed. The second phase of this research concentrates on bridging the anti-spoofing technique of liveliness detection with adaptiveness of biometrics. With the rising demand for involuntary biometric systems, the incorporation of the automatic detection of forgery attacks is becoming very necessary. Adaptability of the system with respect to the change in the trait is another important aspect that this biometric system should be enriched with. Although both liveness and adaptiveness are required to be incorporated in a trusted involuntary biometric system, initial studies in the literature exhibit it as a trade-off. To fulfil the gap, a new user level-based framework for liveness detection is proposed. A variety of contrast, aspect and transform-based quality features are employed to establish liveness. Online classifiers were used to implement adaptiveness of the trait.

Signature and handwriting analysis for real-life application

Real world signature and handwriting verification, as well as recognition scenarios, can be quite challenging pertaining to their topographical scenario. For illustration in a country like in India were multiple official languages are being used officially. More importantly, each of this languages has a different script. Therefore, the influence of script on these scenarios can affect a baseline system. Hence, it will be important to investigate and measure the impact by employing a statistical tool. Consequently, we investigated the impact of the script on multi-script signature and handwriting verification, as well as recognition scenarios by Bhattacharya Distance (BD).

Face analysis in terms of the real-life application scenario

Face analysis including recognition, verification, detection and attribute analysis (such as gender, age estimation) experience additional challenges in real life application scenario. One of the highlighted challenges can be related to biases of attribute analysis depending on it different cross-section of race, age, etc. A Multi-task Convolutional Neural network is proposed employing dynamic joint weight loss to reduce the effect of the aforementioned biasness.

Another prospect of face analysis which can be quite challenging is the detection and analysis of face images from optical phenomenon like a reflection from a glass or shiny wall or for an individual who is present behind a glass plane. Such face images can be used for scenarios where the face image remain uncover in a surveillance scenario.