Telemedicine is drawing greater attention to improve the health care delivery. Video coding being an integral part of any real-time telemedicine system is used to deliver diagnostic video stream to remote physician. Realizing a video coding system customized for telemedicine, using the available technologies poses several challenges. In this paper, we have analyzed state-of-the-art video codecs for adoption in telemedicine customized video conferencing system under low-bandwidth and low-computational scenarios. The experimental results for the selected video codec implementations for medical videos reveal that the HEVC encoder achieves equivalent objective video quality when using approx. 60% bit rate on average. However, the gain in coding efficiency is at the expense of increased computational complexity, which could be dealt with by incorporating adaptive interpolation and selective quality enhancement techniques to achieve real-time performance.
Though slope disaster management practice has accumulated a lot of success and failure of slope treatment project cases, large amounts of data information produced by these engineering cases were not been taken full advantage of therefore caused great waste of resources. Fully development and usage of large amounts of stability-known slope experience undoubtedly have important practical significance. Therefore, based on the usage of previous open-pit mine slope engineering cases, this paper proposes an intelligent slope stability evaluation system based on case-based reasoning. The system uses frame to represent store cases, uses analytic hierarchy process to determine the weight of case attribute, uses similarity calculation method based on Euclidean distance to retrieve case in order to get the result, and finally analyzes the results or reforms the case base if problems exist. Through experimental analysis, this paper proves the validity and feasibility of the system.
Now, gait recognition for identification has received more and more attention from biometrics researchers. Gait Energy Image(GEI) is an efficient represent method and Gabor wavelet has many excellent property, so we use the Gabor wavelet to extract the amplitude and phase feature of GEI, research their recognition ability respectively, at last, fusion the two features in rank level to gait recognition. The algorithm is tested in the CASIA Datasets A and gain high correct recognition rates.
The traditional SIFT algorithm is popular to extract the feature points of target objects, but it also brings the feature points of non-target objects together, leading to mismatching. This paper proposes an improved SIFT algorithm based on adaptive threshold canny operator. In the proposed method, since it has the advantages of accurate edge detection and anti-noise ability, the adaptive threshold canny operator is first employed to detect the edges of an image, and then we can find the feature points by SIFT in the peripheral region of the edges. By introducing the adaptive threshold canny operator, the target objects can be separated from background, largely increasing the matching rate of SIFT algorithm. Experimental results demonstrate that, compared with the traditional SIFT and SURF algorithms, the proposed method can improve the robustness of feature points and further increase the matching rate, meanwhile reducing the cost of running time in a certain extent.
We apply linear sampling method for reconstructing an arbitrarily shaped perfectly conducting crack in limited-view problem. This is based on the physical decomposition of the multi-static response matrix (MSR) in transverse magnetic (TM) polarization. Various numerical results are exhibited to show the strengths and weaknesses of the applied method.
Palm vein identification is a novel biometric technology. Palm vein has high security and very convenient for the user. Many users like this kind of biometric which only use their hands. However, for the NIR light scattering in the skin and difference from different time capture, the recognition performance is not perfect. This paper proposes the algorithm based on partial least squares to extract some directions to compose the classify subspace. The direction change obviously in gray level and has the maximum relationship with the classical information. The coordinate of image in this subspace is used to classification and recognition. Classification is according to the image position in this space. Automation research institute of Chinese academy of sciences database was used to experimental analysis. When the component number is 240, the recognition performance of this scheme reaches the best value: CRR: 98.70%, FAR: 1.30%, FRR: 1.33%. The recognition time is 0.8196s at this component number. Experimental results show that this algorithm improves the recognition performance, suitable for security, attendance, etc, have practical value.
Haze in the environment will hinder the accurate recognition of objects captured in an image. To overcome this problem, image de-hazing processes have been an active technique applied in many research work. Among the available approaches, the one based on the assumption of dark channel prior is able to produce promising results and improved processing speed by integrating the guided filter. However, there are still some limitations existing in this method; particularly the over-range problem makes the appearance of recovered image unnatural. Moreover, its incapability in preserving image brightness frequently requires user intervention. In order to alleviate these shortcomings, the approach presented in this paper is realized through an effective magnitude compression operation. Histogram specification is further exercised for image post-processing. Finally, parameters of both steps are optimized with the particle swarm optimization algorithm. Experiments were conducted with one hundred and thirty hazy images captured in different environmental conditions. Results showed that the proposed method performs better or equivalently in image de-hazing comparing with the approach based on dark channel prior.
Small infrared target detection is one of the key techniques in infrared searching and tracking applications. A novel small target detection method based on the complex filter bank is proposed in this paper. Firstly, a training sample dataset is built and the complex filter bank is utilized to extract the feature representation of each pixel. Then, the support vector machine is used to detect the small target. Finally, a simple adaptive segmentation method is adopted to segment the result image. To evaluate the performance of the proposed method, background clutter suppression and small target detection probability experiments are conducted over real infrared small targets in two image sequences. The experiment results show that the proposed method can not only suppress background clutter effectively, but also detect targets accurately with low false alarm rate.
Traditional vehicle recognition and retrieval systems are almost based on vehicle license plate recognition, which requires the user to provide images containing the license plate of the vehicle. In order to recognize and retrieve vehicles more convenient and efficient, and solve the problem of missing or wrong license plate, this paper proposes a method based on the overall appearance characteristics of the vehicles. The proposed system using a smart phone camera to collect vehicle images and uploads the images to the server through 4G network. The server performs vehicle retrieval and recognition and sends the results back to the phone. To reduce the amount of calculation and improve the accuracy of system, saliency areas of training images are extracted, A hessian-affine detector is used to detect the features of the input image and the SIFT descriptor transforms each feature into a 128-dimensional vector which is finally quantized as "visual word", the ACS (Adjusted Cosine Similarity) and SVM methods are also adopted. The experiments show that the precision of the vehicle recognition is above 96%, and this system has good real-time performance.
Based on the basic principles and characteristics of discrete cosine transform (DCT), we propose a DCT optimization method using genetic algorithm in order to improve the efficiency of image compression. First, we calculate the error between the reconstructed image and its original one. Then, in order to optimize DCT transform matrix coefficients, we use genetic algorithm to find optimal solution by minimizing the mean square error. Finally, we use the optimized transform kernel to process images. Experimental results indicate that the PSNR of the reconstructed images has been improved using the proposed optimization DCT in both cases of image block 4×4 and 8×8.
Pedestrian detection in infrared videos is a task full of potential, and has gotten more and more attention. To robustly detect the pedestrians in infrared image, a PCA-based detecting framework is designed in this paper. The proposed pedestrian detection system can be divided into two parts: training and classification. When the training stage is running, PCA is performed on two different datasets, pedestrian (positive) samples and non-pedestrian (negative) samples, separately. In the classification stage, the system determine whether the input candidate images belong to the positive samples or not by calculating the reconstruction errors for each of them based on the eigenvectors of positive sample and negative sample space. To improve the detecting performance, both the grayscale and edge descriptors are used in the training step. Experimental results indicated that PCA with the combination of grayscale and edge images could achieve the best performance for pedestrian detection.
LBP (Local Binary Pattern) is a commonly used operator to extract LBPH (LBP histogram) of an image for local texture description. For gender classification, we proposed an innovative method by extracting multi-scale LBPH in DoG (Difference of Gaussian) space in this paper. Given a facial image, we firstly preprocess it meticulously to avert the local variations of images which probably be caused by expression, pose and so on. And then we extract multi-scale LBPH features in DoG space which can extract richer local and global interested information of the facial image. Gender classification is performed via a standard binary classifier: SVM. We conducted experiment on 2,410 FERET images and 1,100 images of our collected dataset from the Internet. The best performance of 97.7% is achieved, and the method also performs robustly.
Runway is the most significant sign of airport because of its obvious line and edge features, so traditional methods always use the Hough transform to accomplish the detection. Given the problems that Hough transform needs to consider a lot of factors which spends a lot of time, and there are any breakpoints in results, an improved algorithm based on Hough transform is proposed. First, using the idea of straight-line support range, the range of the phase angle is reduced to a small area to detect, which greatly reducing the amount of computation. Then, the line segments which satisfy certain conditions and have the breakpoints are connected based on the method of line connection. Finally, according to the characteristics of the runway, the unrelated parallel lines are removed, the airport runway is figured out quickly and accurately.
In recent years, the demand for directly visual searching based on images and videos becomes stronger and stronger. Most large-scale image retrieval systems are based on the Bag of Words (BoW) or its variant. For current visual search algorithms which are under the structure of BoW, there are two critical issues should be solved: the quantified visual words may reduce the discriminative power of the local features and the neglect of spatial relationship among local features. To address the problems, we propose a novel method based on saliency and local quadrant constraint. First, we introduce the saliency weights into the quantization stage of BoW. We utilize the image saliency to do inverse document (IDF) weighting. And while generating the histogram expression of image, we count the saliency of features instead of simply counting the number of features. Second, the saliency and the similarity characteristic of deformations in local areas are introduced into our model in the post-processing step to satisfy the constraint in spatial relationship among local features. The operation evaluates the weights of matching features by the saliency of query and candidate images, finds near neighbors of the matching features by a threshold and then estimates whether all the matching features in the local regions follow consistent geometric transformation or not. We evaluate the geometric transformation by the relative quadrant of the matching neighbors to center feature (Local Quadrant Constraint, LQC). Experiments show that the proposed method achieves promising improvement while comparing to other visual search methods. Our methods are well complementary to current visual search methods and give a new idea in modeling the spatial information of image.
The pixels of the retina-like sensor are arranged in different concentric rings, and its output image is in log-polar coordinate. Thus additional residual errors will not be produced when the output image is rotated. Therefore, the retina-like sensor has obvious advantages and wide prospect to be applied in the fields of image rotation and rapid image rotation-elimination. In this paper, a theory concerning the image rotation of the retina-like sensor is proposed, and a solution based on the theory is presented and realized for image rotation-elimination. Several image rotation-elimination experiments have been done, and it turns out that the solution is simply, accurate and rapid.
Researches have shown accent classification can be improved by integrating semantic information into pure acoustic approach. In this work, we combine phonetic knowledge, such as vowels, with enhanced acoustic features to build an improved accent classification system. The classifier is based on Gaussian Mixture Model-Universal Background Model (GMM-UBM), with normalized Perceptual Linear Predictive (PLP) features. The features are further optimized by Principle Component Analysis (PCA) and Hetroscedastic Linear Discriminant Analysis (HLDA). Using 7 major types of accented speech from the Foreign Accented English (FAE) corpus, the system achieves classification accuracy 54% with input test data as short as 20 seconds, which is competitive to the state of the art in this field.
In this paper, we develop a new method to identify Wiener systems. Unlike previous techniques for Winer system identification, our method allows the linear dynamic subsystem to be infinite-impulse response (IIR) and the nonlinear function to be non-differentiable, discontinuous and non-invertible. Two input sequences are designed to estimate the nonlinear function as well as the parameters in the linear system. The convergence analysis is also discussed and the performance of the proposed method is illustrated by simulations.
This paper presents a writing-in-the-air fingertip positioning system using a camera as an input to a computer. The frame difference method is employed to extract the finger movement area and the vertical gradient is calculated in this area to extract the fingertip contour information. Then the fingertip is located by finger geometric properties. 1000-photos library is built to stimulate various operating environments and test the effectiveness of this system. The system achieves a speed of 100 frames per second (fps) with an accuracy rate of 97.0%.
Most classical denoising methods based on wavelet transform will make the edge of an image fuzzy, thus cause the decline of overall effect of image denoising inevitable. Due to this problem, we propose an edge-preserving denoising method with edge detection and probability modelling in this paper. This method applies dual-tree complex wavelet transform to an image and detects the edge of the image based on the wavelet coefficients, thus divides the wavelet coefficients into two parts: the edge part and the non-edge part. For each part, the wavelet coefficients are modelled as a generalized Laplacian distribution, but they are shrinked differently. For the edge part, we preserve more signal information and keep the edge of the image obvious; for the non-edge part, we shrink the wavelet coefficients more sharply to flat the image. Our experimental results, by comparing with several advanced image denoising algorithms, demonstrate that our method can yield better PSNR as well as preserve the edge of the image well.
