Search

Existing face recognition techniques are very successful in recognizing high-resolution facial images. However, their performance is not sufficient on low-resolution facial images. In this thesis we focus on dealing with a typical forensic scenario, where the gallery images are the high-resolution mug-shots in the police database and the probe images are the low-resolution surveillance quality facial images. We proposed a novel method, mixed-resolution biometric comparison, which allows direct low-resolution to high-resolution comparison. The method is based on the likelihood ratio framework where in the derivation of the expression for the likelihood ratio, the combined statistics of the low- and high-resolution images is taken into account. Our experiments on surveillance quality images demonstrate that this method significantly outperforms the state-of-the-art. In literature on low-resolution face recognition, what in some papers is considered as low-resolution, is still considered as high-resolution in other papers. To harmonize the terminology in low-resolution face recognition, we propose a resolution scale. We define the range of low-resolution and further divide it into Upper Low Resolution, Moderately Low Resolution and Very Low Resolution. Because the lack of low-resolution images, most of the existing low-resolution face recognition methods are trained and tested using down-sampled images. In this thesis, we test various face recognition methods and demonstrate that down-sampled images are not fully representative of realistic low-resolution images. We further demonstrate that, inaccurate alignment is the major problem that causes the poor recognition performance on real low-resolution images. In addition, we propose to use matching-score based registration to achieve better alignment and hence better face recognition performance. In conclusion, we propose solutions to compare low-resolution probes with high-resolution galleries which significantly outperform the state-of-the-art on surveillance quality facial images. We emphasise that realistic low-resolution material should be used for training and testing. We focus attention on developing face recognition methods that can actually be useful for real-life applications. We bring an important step forward of low-resolution face recognition for forensic search.

Driver fatigue detection is a significant application in smart cars. In order to improve the accuracy and timeliness of driver fatigue detection, a fatigue detection algorithm based on deeply-learned facial expression analysis is proposed. Specifically, the face key point detection model is first trained by multi block local binary patterns (MB-LBP) and Adaboost classifier. Subsequently, the eyes and mouth state are detected by using the trained model to detect the 24 facial features. Afterwards, we calculate the number of two parameters that can describe the driver's fatigue state and the proportion of the closed eye time within the unit time (PERCLOS) and yawning frequency. Finally, the fuzzy inference system is utilized to deduce the driver's fatigue state (normal, slight fatigue, severe fatigue). Experimental results show that the proposed algorithm can detect driver fatigue degree quickly and accurately.

Recently, more and more researchers are exploring the benefits of super-resolution methods on low-resolution face recognition. However, often results presented are obtained on downsampled high-resolution face images. Because downsampled images are different from real images taken at low resolution, it is important to include real surveillance data. In this paper, we investigate the difference between downsampled images and real surveillance data in two aspects: (1) the influence of resolution on face recognition accuracy, and (2) the improvement of accuracy that can be achieved by super-resolution on these images. Specifically, we will test the following hypotheses: (1) face recognition performance on real images is much worse than on downsampled images, and (2) super-resolution improves the performance of downsampled images more than real images. Our experiments are conducted using videos from the HumanID database. In each video, the target person’s face is captured while he is walking towards the surveillance camera. We detect the faces in the video frames using a Viola-Jones face detector. Then we select face images of four different resolutions: two low-resolution and two high-resolution. The high-resolution images are used for gallery and generating downsampled images. We perform two types of face recognition experiments. In the first type of experiments, three face recognition methods are evaluated for images with different resolution. The three methods are (1) Principal Component Analysis, (2) Linear Discriminant Analysis, and (3) Local Binary Patterns. In the second type, we apply two super-resolution methods: (1) a model based method and (2) a feature based method on the low-resolution (both real and downsampled) images and then compute the face recognition accuracy.

We evaluate the performance of face recognition algorithms on images at various resolutions. Then we show to what extent super-resolution (SR) methods can improve the recognition performance when comparing low-resolution (LR) to high-resolution (HR) facial images. Our experiments use both synthetic data (from the FRGC v1.0 database) and surveillance images (from the SCface database). Three face recognition methods are used, namely Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA) and Local Binary Patterns (LBP). Two SR methods are evaluated. The first method learns the mapping between LR images and the corresponding HR images using a regression model. As a result, the reconstructed SR images are close to the HR images that belong to the same subject and far away from others. The second method compares LR and HR facial images without explicitly constructing SR images. It finds a coherent feature space where the correlation of LR and HR is maximum, and then compute the mapping from LR to HR in this feature space. The performance of the two SR methods are compared to that delivered by the standard face recognition without SR. The results show that LDA is mostly robust to resolution changes while LBP is not suitable for the recognition of LR images. SR methods improve the recognition accuracy when downsampled images are used and the first method provides better results than the second one. However, the improvement for realistic LR surveillance images remains limited.