Abstract: The present disclosure provides an efficient system for cancelable biometric template protection which is not available in the conventional systems. Initially, the system receives a biometric template from a user. The biometric template is one of, a face image, a fingerprint, an iris image and a palmprint. Further, a vector embedding is computed based on the biometric template using a pretrained neural network. After computing the vector embedding, a key based permutation is computed based on the vector embedding and a permutation key. The permutation key is generated based on an Identification number (ID) associated with the user using a random number based key generation technique. Finally, a protected biometric template is generated by computing a permuted embedding corresponding to the biometric template based on the key based permutation using a plurality of fully connected layers pretrained using a weight based center triplet loss function.

Abstract: Image Retrieval is an application of computer vision that deals with searching images in large databases. Conventional methods utilize the entire image to perform the image retrieval task rather than considering specific features. The embodiments herein provide a method and system for feature based image retrieval. Initially, the system receives an input image and a query label. Further, a feature specific encoder is selected from a plurality of feature specific encoders based on the query label. A first set of feature vectors are computed from the input image using the selected feature specific encoder. Further, a Locality Sensitive Hashing (LSH) value is computed from the first set of feature vectors. Finally, a plurality of matching images is obtained from a plurality database images based on a comparison between the computed feature specific LSH value and a plurality of feature specific LSH values stored in a feature specific LSH database.

Abstract: Conventionally, biometric template protection has been achieved to improve matching performance with high levels of security by use of deep convolution neural network models. However, such attempts have prominent security limitations mapping information of images to binary codes is stored in an unprotected form. Given this model and access to the stolen protected templates, the adversary can exploit the False Accept Rate (FAR) of the system. Secondly, once the server system is compromised all the users need to be re-enrolled again. Unlike conventional systems and approaches, present disclosure provides systems and methods that implement encrypted deep neural network(s) for biometric template protection for enrollment and verification wherein the encrypted deep neural network(s) is utilized for mapping feature vectors to a randomly generated binary code and a deep neural network model learnt is encrypted thus achieving security and privacy for data protection.

Abstract: Performance enhancement of face verification systems is credited due to advancement in deep learning methods. However, these systems fail to provide interpretations for decision makings despite their ability to attain high accuracy. Various post-hoc methods have been proposed due to increased demand of deep learning models for better interpretations. However, face verification systems are still prone to adversarial attacks. Present disclosure provides a face verification system and method which addresses the issue of interpretability by employing modular neural network(s), wherein representations for each individual facial feature such as nose, mouth, eyes, etc., are learned separately and verification of input face images is performed. Through experiments, present disclosure demonstrates that the method described herein is resistant to adversarial attacks, thereby addressing another crucial weakness concerning deep learning models.

Abstract: Visual target tracking is task of locating a target in consecutive frame of a video. Conventional systems observe target behavior frames of the video. However, dealing with this problem is very challenging when video has illumination variations, occlusion, change in size and view of the object due to relative motion between camera and object. Embodiments of the present disclosure addresses this problem by implementing Neural Network (NN), its features and their corresponding gradients. Present disclosure explicitly guides the NN by feeding target object of interest (ToI) defined by a bounding box in the first frame of the video. With this guidance, NN generates target activation map via convolutional features map and their gradient maps, thus giving tentative location of the ToI to further exploit to locate target object precisely by using correlation filter(s) and peak location estimator, thus repeating process for every frame of video to track ToI accurately.

Abstract: Image Retrieval is an application of computer vision that deals with searching images in large databases. Conventional methods utilize the entire image to perform the image retrieval task rather than considering specific features. The embodiments herein provide a method and system for feature based image retrieval. Initially, the system receives an input image and a query label. Further, a feature specific encoder is selected from a plurality of feature specific encoders based on the query label. A first set of feature vectors are computed from the input image using the selected feature specific encoder. Further, a Locality Sensitive Hashing (LSH) value is computed from the first set of feature vectors. Finally, a plurality of matching images is obtained from a plurality database images based on a comparison between the computed feature specific LSH value and a plurality of feature specific LSH values stored in a feature specific LSH database.

Abstract: This disclosure relates generally to a method and system for biometric verification. Conventional biometric verification method and system performs one or more computations in non-encrypted domain, thereby leading to security threats. The disclosed method includes performing computations such as enrollment and verification feature vector computation, dimensionality reduction of said feature vectors, and comparison of dimensionally reduced encrypted feature vectors to obtain matching scores indicating the extent of match therebetween between in encrypted domain using fully homomorphic encryption, thereby leading to secure biometric verification.

Abstract: Performance enhancement of face verification systems is credited due to advancement in deep learning methods. However, these systems fail to provide interpretations for decision makings despite their ability to attain high accuracy. Various post-hoc methods have been proposed due to increased demand of deep learning models for better interpretations. However, face verification systems are still prone to adversarial attacks. Present disclosure provides a face verification system and method which addresses the issue of interpretability by employing modular neural network(s), wherein representations for each individual facial feature such as nose, mouth, eyes, etc., are learned separately and verification of input face images is performed. Through experiments, present disclosure demonstrates that the method described herein is resistant to adversarial attacks, thereby addressing another crucial weakness concerning deep learning models.

Abstract: Authentication is a key procedure in information systems. Conventional biometric authentication system is based on a trusted third-party server which is not secure. The present disclosure provides a privacy preserving multifactor biometric authentication for authenticating a client without the third-party authentication server. The server receives a plurality of encrypted biometric features from the client, encrypted using Fully Homomorphic Encryption. Further, the server evaluates the plurality of encrypted biometric features to obtain a client identifier value and a plurality of encrypted resultant values. The server encrypts each of the plurality of resultant values based on a time based nonce and the client identifier value. The encrypted authentication tags and the corresponding resultant values are aggregated by the server and transmitted to the client. The client decrypts the resultant value and the authentication tag and transmits to the server.

Abstract: Authentication is a key procedure in information systems. Conventional biometric authentication system is based on a trusted third-party server which is not secure. The present disclosure provides a privacy preserving multifactor biometric authentication for authenticating a client without the third-party authentication server. The server receives a plurality of encrypted biometric features from the client, encrypted using Fully Homomorphic Encryption. Further, the server evaluates the plurality of encrypted biometric features to obtain a client identifier value and a plurality of encrypted resultant values. The server encrypts each of the plurality of resultant values based on a time based nonce and the client identifier value. The encrypted authentication tags and the corresponding resultant values are aggregated by the server and transmitted to the client. The client decrypts the resultant value and the authentication tag and transmits to the server.

Abstract: Biometric templates (template) are used to store biometric data of one or more users. As biometric data of each user is unique and non-replaceable, template security is a major concern. This disclosure relates generally to template protection, and more particularly to a method and system for biometric template protection. The system generates a final perturbation value for data in the template, based on a first binary code (BC), a second binary code (BC?), feature vectors corresponding to data in the template, and an initial random perturbation. The final perturbation and the feature vectors are mapped to the first binary code. The first binary code represents the template. Every time the first binary code is compromised, the value of first binary code is updated, and the whole process is repeated to generate corresponding final perturbation and then the mapping is done.

Abstract: Conventionally, biometric template protection has been achieved to improve matching performance with high levels of security by use of deep convolution neural network models. However, such attempts have prominent security limitations mapping information of images to binary codes is stored in an unprotected form. Given this model and access to the stolen protected templates, the adversary can exploit the False Accept Rate (FAR) of the system. Secondly, once the server system is compromised all the users need to be re-enrolled again. Unlike conventional systems and approaches, present disclosure provides systems and methods that implement encrypted deep neural network(s) for biometric template protection for enrollment and verification wherein the encrypted deep neural network(s) is utilized for mapping feature vectors to a randomly generated binary code and a deep neural network model learnt is encrypted thus achieving security and privacy for data protection.

Abstract: This disclosure relates generally to a method and system for biometric verification. Conventional biometric verification method and system performs one or more computations in non-encrypted domain, thereby leading to security threats. The disclosed method includes performing computations such as enrollment and verification feature vector computation, dimensionality reduction of said feature vectors, and comparison of dimensionally reduced encrypted feature vectors to obtain matching scores indicating the extent of match therebetween between in encrypted domain using fully homomorphic encryption, thereby leading to secure biometric verification.

Abstract: This disclosure relates generally to a method and system for muting of classified information from an audio using a fuzzy approach. The method comprises converting the received audio signal into text using a speech recognition engine to identify a plurality of classified words from the text to obtain a first set of parameters. Further, a plurality of subwords associated with each classified word are identified to obtain a second set of parameters associated with each subword of corresponding classified word. A relative score is computed for each subword associated with the classified word based on a plurality of similar pairs for the corresponding classified word. A fuzzy muting function is generated using the first set of parameters, the second set of parameters and the relative score associated with each subword. The plurality of subwords associated with each classified word is muted in accordance with the generated fuzzy muting function.

Abstract: Systems and methods for insertion of a watermark into images and tampering detection of the watermarked images by a Convolutional Neural Network (CNN) technique. The traditional systems and methods provide for detecting the tampering of the watermarked images by simply identifying a presence of an inserted watermark into an image but none them provide for inserting a random sequence into input image(s) and then detect the tampering by classifying the input image(s) by a neural network.

Abstract: Target tracking in a video is a highly challenging problem as the target may be effected by its appearance changes along the video, partial occlusions, background clutter, illumination variations, surrounding environment and also due to changes in the motion of the target. Embodiments of the present disclosure address this problem by implementing neural network for convolution feature maps and their gradient maps generation. The proposed two-class neural network (TCNN) is guided by feeding it target of interest defined by a bounding box in a first frame of the video. With this target guidance TCNN generates target activation map by using convolutional features and gradient maps. Target activation map gives tentative location of target, and this is further exploited to locate target precisely by using correlation filter(s) and peak location estimator based on identified context. This process repeats for every frame of the video to track the target accurately.

Abstract: System and method for cheque image data masking are disclosed. In an example, a cheque image and a data file are received, the data file includes data records with values corresponding to sensitive data fields in the cheque image. Further, a template cheque image matching to the cheque image and redacted information associated with the template cheque image are obtained. Furthermore, a blank image snippet is generated for each sensitive data field in the cheque image based on a part of the obtained information about sensitive fields. Moreover, values corresponding to each sensitive field from the input data file are written to the blank image snippet based on the remaining information about sensitive fields. Also, the template cheque image is updated with the image snippet. Data of non-sensitive fields in the cheque image is then copied to the template cheque image, thereby facilitating cheque image data masking.

Abstract: Biometric templates (template) are used to store biometric data of one or more users. As biometric data of each user is unique and non-replaceable, template security is a major concern. This disclosure relates generally to template protection, and more particularly to a method and system for biometric template protection. The system generates a final perturbation value for data in the template, based on a first binary code (BC), a second binary code (BC?), feature vectors corresponding to data in the template, and an initial random perturbation. The final perturbation and the feature vectors are mapped to the first binary code. The first binary code represents the template. Every time the first binary code is compromised, the value of first binary code is updated, and the whole process is repeated to generate corresponding final perturbation and then the mapping is done.

Abstract: Visual target tracking is task of locating a target in consecutive frame of a video. Conventional systems observe target behavior frames of the video. However, dealing with this problem is very challenging when video has illumination variations, occlusion, change in size and view of the object due to relative motion between camera and object. Embodiments of the present disclosure addresses this problem by implementing Neural Network (NN), its features and their corresponding gradients. Present disclosure explicitly guides the NN by feeding target object of interest (ToI) defined by a bounding box in the first frame of the video. With this guidance, NN generates target activation map via convolutional features map and their gradient maps, thus giving tentative location of the ToI to further exploit to locate target object precisely by using correlation filter(s) and peak location estimator, thus repeating process for every frame of video to track ToI accurately.

Abstract: This disclosure relates generally to a method and system for muting of classified information from an audio using a fuzzy approach. The method comprises converting the received audio signal into text using a speech recognition engine to identify a plurality of classified words from the text to obtain a first set of parameters. Further, a plurality of subwords associated with each classified word are identified to obtain a second set of parameters associated with each subword of corresponding classified word. A relative score is computed for each subword associated with the classified word based on a plurality of similar pairs for the corresponding classified word. A fuzzy muting function is generated using the first set of parameters, the second set of parameters and the relative score associated with each subword. The plurality of subwords associated with each classified word is muted in accordance with the generated fuzzy muting function.