Michael Molton

Bio: Michael Molton is an academic researcher from University of Western Australia. The author has contributed to research in topics: Facial recognition system & Upsampling. The author has an hindex of 2, co-authored 4 publications receiving 18 citations.

2D and 3D face recognition using convolutional neural network

Abstract: Face recognition remains a challenge today as recognition performance is strongly affected by variability such as illumination, expressions and poses. In this work we apply Convolutional Neural Networks (CNNs) on the challenging task of both 2D and 3D face recognition. We constructed two CNN models, namely CNN-1 (two convolutional layers) and CNN-2 (one convolutional layer) for testing on 2D and 3D dataset. A comprehensive parametric study of two CNN models on face recognition is represented in which different combinations of activation function, learning rate and filter size are investigated. We find that CNN-2 has a better accuracy performance on both 2D and 3D face recognition. Our experimental results show that an accuracy of 85.15% was accomplished using CNN-2 on depth images with FRGCv2.0 dataset (4950 images with 557 objectives). An accuracy of 95% was achieved using CNN-2 on 2D raw image with the AT&T dataset (400 images with 40 objectives). The results indicate that the proposed CNN model is capable to handle complex information from facial images in different dimensions. These results provide valuable insights into further application of CNN on 3D face recognition.

A Fully Automatic Framework for Prediction of 3D Facial Rejuvenation

Abstract: How will I look afterwards? is a common question asked by the patients undergoing a cosmetic procedure. Cosmetic practitioners at present can only offer subjective and descriptive replies. This subjective prediction is a serious concern for patients undergoing cosmetic treatment and therefore necessitates the development of automatic techniques for facial quantification. This paper proposes a novel machine learning approach to quantify and predict the outcome of 3D facial rejuvenation prior to actual cosmetic procedure. The facial rejuvenation prediction results are achieved by estimating the dermal filler volume in 3D faces. This involves estimation of structural changes in 3D face images and to learn underlying structural mapping. Our preliminary experimental results show that the proposed model achieves superior prediction accuracy on real world dataset compared to baseline methods. The computational time analysis shows that the proposed technique is very efficient (at test time) which makes it suitable for real time applications.

Machine Learning Approaches for Prediction of Facial Rejuvenation Using Real and Synthetic Data

Abstract: This paper proposes a novel machine learning approaches to predict the outcome of facial rejuvenation prior to a cosmetic procedure. This is achieved by estimating the required amount of dermal filler volume that needs to be applied on the face by learning the underlying structural mapping from the pretreatment and posttreatment 3D face images. We develop and train our proposed deep neural network, called Rejuv3DNet, designed specifically to predict the dermal filler volume. We also propose the kernel regression (KR)-based model to validate and improve our volume estimation results using regression. Our other contributions include the development of the first 3D face cosmetic dataset, which consists of real-world pretreatment and posttreatment 3D face images and a novel technique for the generation of synthetic cosmetic treatment 3D face images. Our experimental results show that the proposed Rejuv3DNet and the KR model achieve 62.5% and 66.67%, respectively, on real-world data, while these techniques achieve a prediction accuracy of 75.2% and 89.5%, and 77.2% and 90.1% on our two different synthetic datasets. Our proposed techniques have been found to be computationally efficient, achieving near real-time prediction performance. The reported accuracies are our preliminary results for proof of concept, which can be improved with more data. The proposed approach has the potential for further investigation in the cosmetic surgery domain.

A Training-Free Mesh Upsampling and Morphing Technique for 3D Face Rejuvenation

Abstract: Mesh upsampling and morphing is a challenging problem due to the irregularity and sparseness of the 3D data. Unlike 2D grid of pixels, 3D points do not have any regular structure and spatial order. In this paper, we present an efficient mesh upsampling and morphing technique. The proposed technique does not not require training and does not rely on any particular upsampling model. The key idea is to select and process each mesh triangle based on a heuristic criteria to define the 3D coordinate of a new point. An interactive mesh morphing technique is also introduced to test the effectiveness of the mesh upsampling algorithm. We perform quantitative and qualitative analysis to evaluate the performance of our proposed technique. Our empirical results show that our upsampled points have better uniformity and are located closer to the underlying surfaces. The computational time analysis demonstrates that the proposed technique is very efficient. The average mesh upsampling time is only 2.11sec which makes the proposed technique suitable for real time applications. To further demonstrate the effectiveness of our proposed technique, we evaluate it for a novel task of facial rejuvenation prediction and report our preliminary results in this paper.

3D texture-based face recognition system using fine-tuned deep residual networks.

Abstract: As the technology for 3D photography has developed rapidly in recent years, an enormous amount of 3D images has been produced, one of the directions of research for which is face recognition. Improving the accuracy of a number of data is crucial in 3D face recognition problems. Traditional machine learning methods can be used to recognize 3D faces, but the face recognition rate has declined rapidly with the increasing number of 3D images. As a result, classifying large amounts of 3D image data is time-consuming, expensive, and inefficient. The deep learning methods have become the focus of attention in the 3D face recognition research. In our experiment, the end-to-end face recognition system based on 3D face texture is proposed, combining the geometric invariants, histogram of oriented gradients and the fine-tuned residual neural networks. The research shows that when the performance is evaluated by the FRGC-v2 dataset, as the fine-tuned ResNet deep neural network layers are increased, the best Top-1 accuracy is up to 98.26% and the Top-2 accuracy is 99.40%. The framework proposed costs less iterations than traditional methods. The analysis suggests that a large number of 3D face data by the proposed recognition framework could significantly improve recognition decisions in realistic 3D face scenarios.

A Fully Automatic Framework for Prediction of 3D Facial Rejuvenation

Abstract: How will I look afterwards? is a common question asked by the patients undergoing a cosmetic procedure. Cosmetic practitioners at present can only offer subjective and descriptive replies. This subjective prediction is a serious concern for patients undergoing cosmetic treatment and therefore necessitates the development of automatic techniques for facial quantification. This paper proposes a novel machine learning approach to quantify and predict the outcome of 3D facial rejuvenation prior to actual cosmetic procedure. The facial rejuvenation prediction results are achieved by estimating the dermal filler volume in 3D faces. This involves estimation of structural changes in 3D face images and to learn underlying structural mapping. Our preliminary experimental results show that the proposed model achieves superior prediction accuracy on real world dataset compared to baseline methods. The computational time analysis shows that the proposed technique is very efficient (at test time) which makes it suitable for real time applications.

Smart drone with real time face recognition

Abstract: This paper develops a solution to help the task force using a face recognition based Unmanned Aerial Vehicle (UAV) to identify the criminals, missing people, civilians and for surveillance. A facial recognition system is a technology which involves the understanding of how the faces are detected and recognized, usually employed to authenticate users through Identity verification services and computing facial features from a given image. This work elucidates an unmanned aerial vehicle with a camera attached that is linked to Face Recognition software and used to control a robot through a wireless remote. When the drone's camera angle is within 37 degrees, the accuracy is 98.6%, indicating that the suggested method can handle changing drone placements. Some applications are to assist the police officials with their investigations like search and rescue, investigations of incidents, crime analysis and crowd monitoring.

Spatial hierarchical analysis deep neural network for RGB-D object recognition

Abstract: Deep learning based object recognition methods have achieved unprecedented success in the recent years. However, this level of success is yet to be achieved on multimodal RGB-D images. The latter can play an important role in several computer vision and robotics applications. In this paper, we present spatial hierarchical analysis deep neural network, called ShaNet, for RGB-D object recognition. Our network consists of convolutional neural network (CNN) and recurrent neural network (RNNs) to analyse and learn distinctive and translationally invariant features in a hierarchical fashion. Unlike existing methods, which employ pre-trained models or rely on transfer learning, our proposed network is trained from scratch on RGB-D data. The proposed model has been tested on two different publicly available RGB-D datasets including Washington RGB-D and 2D3D object dataset. Our experimental results show that the proposed deep neural network achieves superior performance compared to existing RGB-D object recognition methods.

Machine Learning Approaches for Prediction of Facial Rejuvenation Using Real and Synthetic Data

Abstract: This paper proposes a novel machine learning approaches to predict the outcome of facial rejuvenation prior to a cosmetic procedure. This is achieved by estimating the required amount of dermal filler volume that needs to be applied on the face by learning the underlying structural mapping from the pretreatment and posttreatment 3D face images. We develop and train our proposed deep neural network, called Rejuv3DNet, designed specifically to predict the dermal filler volume. We also propose the kernel regression (KR)-based model to validate and improve our volume estimation results using regression. Our other contributions include the development of the first 3D face cosmetic dataset, which consists of real-world pretreatment and posttreatment 3D face images and a novel technique for the generation of synthetic cosmetic treatment 3D face images. Our experimental results show that the proposed Rejuv3DNet and the KR model achieve 62.5% and 66.67%, respectively, on real-world data, while these techniques achieve a prediction accuracy of 75.2% and 89.5%, and 77.2% and 90.1% on our two different synthetic datasets. Our proposed techniques have been found to be computationally efficient, achieving near real-time prediction performance. The reported accuracies are our preliminary results for proof of concept, which can be improved with more data. The proposed approach has the potential for further investigation in the cosmetic surgery domain.