Earticle

Automatic classification of analog and digital modulation signals plays an important role in communication application such as an intelligent demodulator, interference identification and monitoring. The automatic recognition of the modulation format of a detected signal, the intermediate step between signal detection and demodulation, is a major task of an intelligent receiver, with various civilian and military applications. This paper presents a new approach for automatic modulation classification for digitally modulated signals. This method utilizes a signal representation known as the modulation model. The modulation model provides a signal representation that is convenient for subsequent analysis, such as estimating modulation parameters. The modulation parameters to be estimated are the carrier frequency, modulation type, and bit rate. The modulation model is formed via autoregressive spectrum modeling. The modulation model uses the instantaneous frequency and bandwidth parameters as obtained from the roots of the autoregressive polynomial. This method is also classifies accurately under low carrier to noise ratio (CNR). This paper is also presents an improved version of S-Transform for time frequency analysis of different digitally modulated signals to observe variations of amplitude, frequency and phase.

Forest height is important information for many forest management activities and is a critical parameter in models of ecosystem procedures. Recently, there have been plenty of researches on the retrieval of forest height by single baseline PolInSAR such as the ESPRIT method, three-stage inversion but the methods tend to underestimate the forest height due to attenuation of the electromagnetic waves in the ground medium and vary widely in their sensitivities. This paper proposes a novel algorithm to retrieve forest height using an adaptive scattering model-based decomposition technique with PolInSAR data. The object is to describe each interferometry cross correlation as a sum of contributions corresponding to odd bounce, double bounce and volume scattering processes. This algorithm enables the retrieval not only of the vegetation parameters but also of the magnitude associated with each mechanism. Another advantage of the proposed algorithm is that it makes use of all the information provided by the covariance matrix, which remains unachieved in the previous model-based decompositions. The proposed algorithm has been tested with simulated data from PolSARProSim software and spaceborne data from a test site. Experimental results indicate that accuracy of the forest height estimation can be enhanced by the proposed algorithm.

Several speaker identification systems are giving good performance with clean speech but are affected by the degradations introduced by noisy audio conditions. To deal with this problem, we investigate the use of complementary information at different levels for computing a combined match score for the unknown speaker. In this work, we observe the effect of two supervised machine learning approaches including support vectors machines (SVM) and naïve bayes (NB). We define two feature vector sets based on mel frequency cepstral coefficients (MFCC) and relative spectral perceptual linear predictive coefficients (RASTA-PLP). Each feature is modeled using the Gaussian Mixture Model (GMM). Several ways of combining these information sources give significant improvements in a text-independent speaker identification task using a very large telephone degraded NTIMIT database.

Human object extraction from infrared image has broad applications, and has become an active research area in image processing community. Combined with chaos differential evolution (CDE) algorithm and morphological operators, a novel infrared human target extraction method is proposed based on nonextensive fuzzy entropy. Firstly, the image was transformed into a fuzzy domain by fuzzy membership function, and the image nonextensive fuzzy entropy was constructed. Then, the image was segmented by thresholding based on the maximum entropy principle and the pseudoadditivity rule of nonextensive entropy. In order to reduce the search time of optimal threshold selection, the CDE algorithm was presented. Finally, the object was extracted using morphological operators to denoise, fill cavity on the threshold segmented image. Experimental results show that the proposed method is efficient and requires less computation time.

The matching accuracy of remote sensing image-pair has a great relationship with the brightness value of the corresponding points. Generally the corresponding points of remote image-pair have great brightness difference, which may affect and reduce the matching precision. Therefore, in order to acquire high-precision stereo matching, a new method was proposed in this paper. This method can balance the brightness by combining the local information and middle axis cumulative histogram of image-pair and can adjust the brightness difference to be the same between image-pair. Both theoretical analysis and experimental results show that this new method can achieve better result in balancing brightness difference, keeping details and improving matching accuracy of remote sensing stereo image-pair effectively.

In the past decades, many nonlinear partial differential equation (PDE) based denosing methods have been suggested, among which the curvature-preserving PDE image denosing method is one of the outstanding models. To effectively preserve image edge well, a tensor driven linear integral convolutions based Image Denosing Method is proposed, which employ total variation flow based nonlinear structure tensor to analysis different integral curve. It is a new implementation of our former work [10]. Experimental results show that the new method can achieve better denosing results in a variety of standard test images, and the new approach shows superior performance on edge and curvature preserving face image and texture image.

Data fusion aims at synergistic use of information and knowledge from different sources to aid in the overall understanding of a phenomenon. In the domain of remote sensing, where images are acquired by multiple sources or by the same source in multiple acquisition contexts, the data made available by different sources are complementary to each other, proper fusion of the data can bring better and consistent interpretation of the scene. The paper presents application of Kalman filter at pixel-level fusion. The input data collected from Ozone Monitoring Instrument (OMI) on NASA’s Aura satellite is subjected to the proposed algorithm. The performance of the algorithm is evaluated by few well-known image quality metrics.

With recent advances in Internet computing and its growing importance in our day to day life, the need for confidential communication has increased. The falling of sensitive and confidential data such as top intelligence secrets into an enemy nation’s hands can lead to the misuse of technology and destruction of masses. Therefore, researchers have developed techniques such as cryptography and steganography to protect the data being transmitted. Cryptography uses mathematical algorithms to convert the data into an unreadable form and steganography on the other hand hides the data in a carrier such as image, data, audio or video. Both of these algorithms have disadvantages i.e. in case of cryptography the altered text can be easily detected by an intruder while in steganography the distortion of the carrier can attract an adversary. The proposal combines the advantages of both the techniques and provides a study tool for security. It calls for the combination of the two techniques which increases the brute force search time but at the same time ensures that the increase in time complexity of the result and process is within acceptable limits. Moreover this paper depicts a mathematical calculation of the errors in the embedded image with respect to the original image through the calculation of mean square error, peak signal to noise ratio and histogram representations.

A new method for head posture and gaze direction estimation is proposed. Firstly, three models of head position are established and postures are judged based on triangle attribute constituted by eyes and mouth. Then pupil is located using Hough transform in eye area. With the method of horizontal and vertical projection and eye prior knowledge, the normal eye outline is fitted. Finally, gaze direction is estimated according to the position of pupil in normal state eye and head posture. The experimental results demonstrate the proposed method can accurately detect head posture and gaze direction. For considering head posture, the method has more accuracy in gaze estimation.

We present a system for acquiring synchronized multi-view color and depth (RGB-D) data using multiple off-the-shelf Microsoft Kinect and a new method for reconstructing spatio-temporally coherent 3D animation from time-varying dynamic RGB-D data. Our acquisition system is independent of any specific hardware component for the synchronization of the camera system. We show that the data acquired by our framework can be synchronously registered in a global coordinate system and then can be used to reconstruct the 3D animation of a dynamic scene. The main benefit of our work is that instead of relying on expensive multi-view video capture setups, multiple low cost Microsoft Kinect sensors can capture both the image and the depth data to do a 360o reconstruction of a dynamic scene. We also present a new algorithm for tracking dynamic three-dimensional point cloud data that can be used to reconstruct a time-coherent representation of a 3D animation without using any template model or a-prior assumption about the underlying surface. We show that despite some limitations imposed by the hardware for the synchronous acquisition of the data we can get reasonably good reconstruction of the animated 3D geometry, which can be used in a number of applications.

A new switching median filter is proposed for denoising of gray-scale images, extremely corrupted by salt-and-pepper noise. The proposed model for noise removal is a multiscale detection based adaptive median filter. This method consists of mainly two parts, namely, the thresholding based multiscale noise detection and the filtering. The detection of impulse noise is carried out in two stages. First, multiscale filtering of the corrupted image is carried out using Gaussian kernels at different scales and errors between the original and the filtered images at different scales are obtained. In the next stage, the errors at different scales are added and then thresholded to detect the impulse noise. The filtering of impulses, detected in the first stage of the proposed filter, is finally carried out using an adaptive median filter. Incorporation of a multiscale method into the noise detection stage followed by thresholding has led to more reliable and efficient impulse noise detection, especially, at high noise ratios. To validate the efficacy of proposed scheme, extensive simulations and comparisons are done with the competent schemes under a wide range (10% to 90%) of noise densities. The results show that the proposed scheme works much better in suppressing high level noise than other schemes, keeping the edges and fine details of the original image almost intact.

Remote sensing image registration is the basis of remote sensing image mosaic and fusion, while extracting appropriate control point pairs have great significance to establish the mapping relationship for registration of two remote sensing images. An automatic extracting method of control point pairs is proposed. Firstly, it determines the common areas of two images and divides them into blocks uniformly. The corresponding blocks of the two images are marked with the same sequence number to establish the one-to-one regional matching relation. Then, for each block, the multi-scale Harris corner detection method is adopted to detect image features which are described by feature descriptors of SIFT algorithm. Finally the regional matching strategy is performed to match the features. Experiments show that the proposed method can extract control point pairs with reasonable distribution and high precision, and these point pairs conduct to improve the precision of remote sensing image matching.

Keypoint selection is the important step in object recognition, including general object classification, human tracing and human pose discrimination etc .This paper proposes a more accurate modified key point selection algorithm by modifying SIFT in the stage of extreme point selection. In machine vision or computer vision, including human pose recognition, to select key points, the traditional SIFT completes this according to the extremes derived from LoG (Laplacian of Gaussian) convolution with image, which provides scale invariance features for key points. The extreme points’ position is the foundation of feature descriptor for the gradient calculation in the next step. But in the process of images convoluting with the difference of Gaussian function to attain the extreme point, bias is produced because the extreme points’ positions aren’t accurate. We modify the extreme points’ selection to make key points more accurate with less bias to the theoretical points. Simulation with about 3500 images of different resolutions gives the AIPR (adjusted interest point ratio) and illustrates the universality of extreme points’ selection and verifies the values of this algorithm.

Texture contains high and low frequency information which could be hierarchically extracted by scattering the texture along multiple paths, with a cascade of wavelet modulus operators implemented in a deep convolutional network, which builds a scattering energy distribution network. Therefore, the scattering transform is used, in this paper, to get texture energy features. Besides, the classification of scattering energy feature matrix at all levels is done by using the Ostu global threshold processing method. Experimental results indicate that high accuracy can be achieved for both texture segmentation and license plate location with the proposed methods.

Vehicle model recognition from images is one of challenging fields for supporting intelligence transport system. Existing methods only deal with vehicle in fixed view. However, in video, rotation is occurred and it decreases performance of recognition. To overcome this problem, rectification method about skew though rotated view is needed. We employ the robust license plate detection method using straight line template matching to get ROI. Also, we perform recognition process toward vehicle in rotated view whose angle is 0° to 15° after proposed image rectification from proposed rectification method.