Shengzhe Li

Fast and Accurate Rigid Registration of 3D CT Images by Combining Feature and Intensity

Computed tomography (CT) images are widely used for the analysis of the temporal evaluation or monitoring of the progression of a disease. The follow-up examinations of CT scan images of the same patient require a 3D registration technique. In this paper, an automatic and robust registration is proposed for the rigid registration of 3D CT images. The proposed method involves two steps. Firstly, the two CT volumes are aligned based on their principal axes, and then, the alignment from the previous step is refined by the optimization of the similarity score of the image’s voxel. Normalized cross correlation (NCC) is used as a similarity metric and a downhill simplex method is employed to find out the optimal score. The performance of the algorithm is evaluated on phantom images and knee synthetic CT images. By the extraction of the initial transformation parameters with principal axis of the binary volumes, the searching space to find out the parameters is reduced in the optimization step. Thus, the overall registration time is algorithmically decreased without the deterioration of the accuracy. The preliminary experimental results of the study demonstrate that the proposed method can be applied to rigid registration problems of real patient images.

Fingerprint liveness detection based on multiple image quality features

Recent studies have shown that the conventional fingerprint recognition systems are vulnerable to fake attacks, and there are many existing systems that need to update their anti-spoofing capability inexpensively. This paper proposes an image quality-based fake detection method to address this problem. Three effective fake/live quality measures, spectral band energy, middle ridge line and middle valley line, are extracted firstly, and then, these features are fused and tested on a fake/live dataset using SVM and QDA classifiers. Experimental results demonstrate that the proposed method is promising in increasing the security of the existing fingerprint authentication system by only updating the software.

Real-Time Human Action Recognition Using CNN Over Temporal Images for Static Video Surveillance Cameras

This paper proposes a real-time human action recognition approach to static video surveillance systems. This approach predicts human actions using temporal images and convolutional neural networks CNN. CNN is a type of deep learning model that can automatically learn features from training videos. Although the state-of-the-art methods have shown high accuracy, they consume a lot of computational resources. Another problem is that many methods assume that exact knowledge of human positions. Moreover, most of the current methods build complex handcrafted features for specific classifiers. Therefore, these kinds of methods are difficult to apply in real-world applications. In this paper, a novel CNN model based on temporal images and a hierarchical action structure is developed for real-time human action recognition. The hierarchical action structure includes three levels: action layer, motion layer, and posture layer. The top layer represents subtle actions; the bottom layer represents posture. Each layer contains one CNN, which means that this model has three CNNs working together; layers are combined to represent many different kinds of action with a large degree of freedom. The developed approach was implemented and achieved superior performance for the ICVL action dataset; the algorithm can run at around 20 frames per second.

Assessing the level of difficulty of fingerprint datasets based on relative quality measures

Understanding the difficulty of a dataset is of primary importance when it comes to testing and evaluating fingerprint recognition systems or algorithms because the evaluation result is dependent on the dataset. The difficulty exhibited in this paper represents how difficult it is to achieve better recognition accuracy within the specific dataset. Proposed in this paper is a general framework for assessing the level of difficulty of fingerprint datasets based on quantitative measurements of not only the sample quality of individual fingerprints but also the relative differences between genuine pairs, such as common area and deformation. The experimental results over various datasets demonstrate that the proposed method can predict the level of difficulty of fingerprint datasets which coincide with the equal error rates produced by four comparison algorithms. The proposed method is independent of comparison algorithms and can be performed automatically.

Assessing the Difficulty Level of Fingerprint Datasets Based on Relative Quality Measures

Understanding the difficulty of a dataset is of primary importance when it comes to testing and evaluating fingerprint recognition systems or algorithms because the evaluation result is dependent on the dataset. Proposed in this paper is a general framework of assessing the level of difficulty of fingerprint datasets based on quantitative measurements of not only the sample quality of individual fingerprints but also relative differences between genuine pairs, such as common area and deformation. The experimental results over multi-year FVC datasets demonstrate that the proposed method can predict the relative difficulty levels of the fingerprint datasets which coincide with the equal error rates produced by two matching algorithms. The proposed framework is independent of matching algorithms and can be performed automatically.

Noise removal for multi-echo MR images using global enhancement

Magnetic Resonance Images are corrupted by random noise which affects the accuracy of the quantitative measurements when acquiring the data. This paper proposes an effective noise removal method for multiple-echo MR images using global enhancement. Applying inverse histogram equalization to the mean image, not only the noise is removed from the MRI but also the noise-free image can be reconstructed. Firstly, the background is segmented using morphological operations and curve fitting based on histogram equalization is performed to normalize the noise signal. Then, the average filtering is applied to the segmented MR images in order to remove the noise. Next, the noise-removed multi-echo MR images are reconstructed using the inverse histogram equalization. The experimental results demonstrate that the proposed method can eliminate not only anisotropic noises but also the flow artifacts in multi-echo MR Images.

Robust fire detection using logistic regression and randomness testing for real-time video surveillance

This paper proposes a real-time fire detection algorithm for video surveillance. Firstly, candidate fire regions (CFRs) are detected using modified conventional methods, that is, the detection of moving regions and fire-colored pixels. In order to avoid false alarms, effective color and shape-based features are extracted from CFRs. Then, the set of features are fed into the logistic regression to classify the fire and non-fire regions. A randomness test over the features is further adopted for the final fire verification. Experimental results show that the proposed approach is more robust and fast. False alarms due to ordinary motion of flame colored moving objects are reduced with a great amount, compared to the existing video based fire detection systems.

Improvement of Accuracy for Human Action Recognition by Histogram of Changing Points and Average Speed Descriptors

Human action recognition has become an important research topic in computer vision area recently due to many applications in the real world, such as video surveillance, video retrieval, video analysis, and human-computer interaction. The goal of this paper is to evaluate descriptors which have recently been used in action recognition, namely Histogram of Oriented Gradient (HOG) and Histogram of Optical Flow (HOF). This paper also proposes new descriptors to represent the change of points within each part of a human body, caused by actions named as Histogram of Changing Points (HCP) and so-called Average Speed (AS) which measures the average speed of actions. The descriptors are combined to build a strong descriptor to represent human actions by modeling the information about appearance, local motion, and changes on each part of the body, as well as motion speed. The effectiveness of these new descriptors is evaluated in the experiments on KTH and Hollywood datasets. Category: Smart and intelligent computing

Type-Independent Pixel-Level Alignment Point Detection for Fingerprints

Robust alignment point detection is still a challenging problem in fingerprint recognition, especially for arch type fingerprints. Proposed in this paper is a method of detecting a pixel-level alignment point from mated fingerprints regardless of the type based on pixel-level orientation field. Given a fingerprint, firstly, pixel-level orientation field is computed using multi-scale Gaussian filtering. Secondly, a vertical symmetry line is extracted from the orientation field, based on which the fingerprint type is classified, either arch or non-arch type. For non-arch mated pairs, the pixel-level singular points (core or delta) are adopted as candidate alignment points and be verified by point-pattern matching and the average orientation difference between the orientation fields. And, for arch mated pairs, the alignment points are detected at the maximum in the angular difference and the orientation certainty level over the symmetry lines. The proposed method is tested over the FVC 2000 DB2a, and 95.93% mated fingerprint pairs are aligned within one ridge-width displacement.

Detecting and visualizing cartilage thickness without a shape model

This paper proposes a cartilage thickness detection and visualization method that does not utilize a shape model. The proposed method consists of three parts: volume of interest (VOI) initialization, bone segmentation, and cartilage thickness visualization. For VOI initialization, a novel 3D U-shape cuboidal filter is proposed to detect individual bones such as the femur, tibia, and patella, and for bone segmentation, a hybrid level-set method is adopted. Finally, a surface normal based approach is presented for measuring and visualizing the cartilage thickness. The advantage of the proposed method compared to other methods is that it does not require a shape model or any training process. The results demonstrate that the proposed method can be used for inspecting cartilage damage and loss.