A GAUSSIAN MIXTURE MODEL-BASED REGULARIZATION METHOD IN ADAPTIVE IMAGE RESTORATION

A GMM (Gaussian Mixture Model) based adaptive image restoration is proposed in this paper. The feature vectors of pixels are selected and extracted. Pixels are clustered into smooth,edge or detail texture region according to variance-sum criteria function of the feature vectors. Then pa-rameters of GMM are calculated by using the statistical information of these feature vectors. GMM predicts the regularization parameter for each pixel adaptively. Hopfield Neural Network (Hopfield-NN) is used to optimize the objective function of image restoration,and network weight value matrix is updated by the output of GMM. Since GMM is used,the regularization parameters share properties of different kind of regions. In addition,the regularization parameters are different from pixel to pixel. GMM-based regularization method is consistent with human visual system,and it has strong gener-alization capability. Comparing with non-adaptive and some adaptive image restoration algorithms,experimental results show that the proposed algorithm obtains more preferable restored images.

Adaptive Image Zooming Algorithm Combining Total Variation Minimization and a Second-order Functional

An image zooming algorithm by using partial differential equations(PDEs) is proposed here. It combines the second-order PDE with a fourth-order PDE. The combined algorithm is able to preserve edges and at the same time avoid the blurry effect in smooth regions. An adaptive function is used to combine the two PDEs. Numerical experiments illustrate advantages of the proposed model.

Multi-scale traffic vehicle detection based on faster ReCNN with NAS optimization and feature enrichment

It well known that vehicle detection is an important component of the field of object detection.However,the environment of vehicle detection is particularly sophisticated in practical processes.It is comparatively difficult to detect vehicles of various scales in traffic scene images,because the vehicles partially obscured by green belts,roadblocks or other vehicles,as well as influence of some low illumination weather.In this paper,we present a model based on Faster ReCNN with NAS optimization and feature enrichment to realize the effective detection of multi-scale vehicle targets in traffic scenes.First,we proposed a Retinex-based image adaptive correction algorithm(RIAC)to enhance the traffic images in the dataset to reduce the influence of shadow and illumination,and improve the image quality.Second,in order to improve the feature expression of the backbone network,we conducted Neural Architecture Search(NAS)on the backbone network used for feature extraction of Faster ReCNN to generate the optimal cross-layer connection to extract multi-layer features more effectively.Third,we used the object Feature Enrichment that combines the multi-layer feature information and the context information of the last layer after cross-layer connection to enrich the information of vehicle targets,and improve the robustness of the model for challenging targets such as small scale and severe occlusion.In the implementation of the model,K-means clustering algorithm was used to select the suitable anchor size for our dataset to improve the convergence speed of the model.Our model has been trained and tested on the UN-DETRAC dataset,and the obtained results indicate that our method has art-of-state detection performance.

Detection with High Resolution Radar: Advanced Topics & Potential Applications

In a previous companion paper [1], the potential advantages of high resolution radar for improved target detection were introduced. In particular, the concept of shaping both the transmitted waveform and the receiving processor in accordance to the expected target down-range profile was highlighted and performance predictions were provided. In this paper, we present and evaluate an adaptive scheme devised to on-line estimate the target profile, in order to overcome a limited a-priori knowledge. In addition, we introduce a more general model of target impulse response, based on a statistical description, and we discuss the corresponding processing scheme and detection performance.

An amplitude-preserved adaptive focused beam seismic migration method

Gaussian beam migration （GBM） is an effec- tive and robust depth seismic imaging method, which overcomes the disadvantage of Kirchhoff migration in imaging multiple arrivals and has no steep-dip limitation of one-way wave equation migration. However, its imaging quality depends on the initial beam parameters, which can make the beam width increase and wave-front spread with the propagation of the central ray, resulting in poor migration accuracy at depth, especially for exploration areas with complex geological structures. To address this problem, we present an adaptive focused beam method for shot-domain prestack depth migration. Using the infor- mation of the input smooth velocity field, we first derive an adaptive focused parameter, which makes a seismic beam focused along the whole central ray to enhance the wave- field construction accuracy in both the shallow and deep regions. Then we introduce this parameter into the GBM, which not only improves imaging quality of deep reflectors but also makes the shallow small-scale geological struc- tures well-defined. As well, using the amplitude-preserved extrapolation operator and deconvolution imaging condi- tion, the concept of amplitude-preserved imaging has been included in our method. Typical numerical examples and the field data processing results demonstrate the validity and adaptability of our method.

Adaptive optics scanning laser ophthalmoscopy in fundus imaging, a review and update

Adaptive optics scanning laser ophthalmoscopy（AOSLO） has been a promising technique in funds imaging with growing popularity. This review firstly gives a brief history of adaptive optics（AO） and AO-SLO. Then it compares AO-SLO with conventional imaging methods（fundus fluorescein angiography, fundus autofluorescence, indocyanine green angiography and optical coherence tomography） and other AO techniques（adaptive optics flood-illumination ophthalmoscopy and adaptive optics optical coherence tomography）. Furthermore, an update of current research situation in AO-SLO is made based on different fundus structures as photoreceptors（cones and rods）, fundus vessels, retinal pigment epithelium layer, retinal nerve fiber layer, ganglion cell layer and lamina cribrosa. Finally, this review indicates possible research directions of AO-SLO in future.

Stochastic parallel gradient descent based adaptive optics used for a high contrast imaging coronagraph

An adaptive optics （AO） system based on a stochastic parallel gradient descent （SPGD） algorithm is proposed to reduce the speckle noises in the optical system of a stellar coronagraph in order to further improve the contrast. The principle of the SPGD algorithm is described briefly and a metric suitable for point source imaging optimization is given. The feasibility and good performance of the SPGD algorithm is demonstrated by an experimental system featured with a 140-actuator deformable mirror and a Hartrnann-Shark wavefront sensor. Then the SPGD based AO is applied to a liquid crystal array （LCA） based coronagraph to improve the contrast. The LCA can modulate the incoming light to generate a pupil apodization mask of any pattern. A circular stepped pattern is used in our preliminary experiment and the image contrast shows improvement from 10^-3 to 10^-4.5 at an angular distance of 2A/D after being corrected by SPGD based AO.

Determining the imaging plane of a retinal capillary layer in adaptive optical imaging

Even in the early stage,endocrine metabolism disease may lead to micro aneurysms in retinal capillaries whose diameters are less than 10 μm.However,the fundus cameras used in clinic diagnosis can only obtain images of vessels larger than 20 μm in diameter.The human retina is a thin and multiple layer tissue,and the layer of capillaries less than10 μm in diameter only exists in the inner nuclear layer.The layer thickness of capillaries less than 10 μm in diameter is about 40 μm and the distance range to rod＆cone cell surface is tens of micrometers,which varies from person to person.Therefore,determining reasonable capillary layer（CL） position in different human eyes is very difficult.In this paper,we propose a method to determine the position of retinal CL based on the rod＆cone cell layer.The public positions of CL are recognized with 15 subjects from 40 to 59 years old,and the imaging planes of CL are calculated by the effective focal length of the human eye.High resolution retinal capillary imaging results obtained from 17 subjects with a liquid crystal adaptive optics system（LCAOS） validate our method.All of the subjects＇ CLs have public positions from 127 μm to 147 μm from the rod＆cone cell layer,which is influenced by the depth of focus.

Multifeature Statistical TV Tracker

A multifeature statistical image segmentation algorithm is described. Multiple features such as grey, edge magnitude and correlation are combined to form a multidimensional space statistics. The statistical algorithm is used to segment an image using the decision curved surface determined by the multidimensional feature function. The segmentation problem which is difficult to solve using the features independently will be readily solved using the same features jointly. An adaptive segmentation algorithm is discussed. Test results of the real-time TV tracker newly developed have shown that the segmentation algorithm discussed here improves effectively the image segmentation quality and system tracking performance.

Fast Adaptive Wavelet for Remote Sensing Image Compression

Remote sensing images are hard to achieve high compression ratio because of their rich texture. By analyzing the influence of wavelet properties on image compression, this paper proposes wavelet construction rules and builds a new biorthogonal wavelet construction model with parameters. The model parameters are optimized by using genetic algorithm and adopting energy compaction as the optimization object function. In addition, in order to resolve the computation complexity problem of online construction, according to the image classification rule proposed in this paper we construct wavelets for different classes of images and implement the fast adaptive wavelet selection algorithm （FAWS）. Experimental results show wavelet bases of FAWS gain better compression performance than Daubechies9/7.

Face Recognition via Adaptive Image Combination

Dimension reduction and manifold learning are the two most popular feature extraction methods.The two methods focus on spatial locality as a guiding principle to find a low-dimensional basis for describing high-dimensional data,but no bases or features are more spatially localized than the original image pixels.So,adaptive image combination is presented to represent a class by a combined sample.The combined sample is a linear combination of original samples in the same class.Adaptive image combination (AIC) find the best combination coefficients by minimizing the intrapersonal distance and maximizing the interpersonal distance.Experimental results show that AIC is effective.

Real-time adaptive particle image velocimetry for accurate unsteady flow field measurements

Almost all conventional open-loop particle image velocimetry(PIV) methods employ fixed-interval-time optical imaging technology and the time-consuming cross-correlation-based PIV measurement algorithm to calculate the velocity field.In this study,a novel real-time adaptive particle image velocity(RTA-PIV) method is proposed to accurately measure the instantaneous velocity field of an unsteady flow field.In the proposed closed-loop RTA-PIV method,a new correlation-filter-based PIV measurement algorithm is introduced to calculate the velocity field in real time.Then,a Kalman predictor model is established to predict the velocity of the next time instant and a suitable interval time can be determined.To adaptively adjust the interval time for capturing two particle images,a new high-speed frame-straddling vision system is developed for the proposed RTA-PIV method.To fully analyze the performance of the RTA-PIV method,we conducted a series of numerical experiments on ground-truth image pairs and on real-world image sequences.

An adaptive image sparse reconstruction method combined with nonlocal similarity and cosparsity for mixed Gaussian-Poisson noise removal

Compressed sensing(CS) has achieved great success in single noise removal. However, it cannot restore the images contaminated with mixed noise efficiently. This paper introduces nonlocal similarity and cosparsity inspired by compressed sensing to overcome the difficulties in mixed noise removal, in which nonlocal similarity explores the signal sparsity from similar patches, and cosparsity assumes that the signal is sparse after a possibly redundant transform. Meanwhile, an adaptive scheme is designed to keep the balance between mixed noise removal and detail preservation based on local variance. Finally, IRLSM and RACoSaMP are adopted to solve the objective function. Experimental results demonstrate that the proposed method is superior to conventional CS methods, like K-SVD and state-of-art method nonlocally centralized sparse representation(NCSR), in terms of both visual results and quantitative measures.

An improved adaptive preprocessing method for TDI CCD images

In order to achieve high quality images with time-delayed integration(TDI) charge-coupled device(CCD) imaging system, an improved adaptive preprocessing method is proposed with functions of both denoising and edge enhancement. It is a weighted average filter integrating the average filter and the improved range filter. The weighted factors are deduced in terms of a cost function, which are adjustable to different images. To validate the proposed method, extensive tests are carried out on a developed TDI CCD imaging system. The experimental results confirm that this preprocessing method can fulfill the noise removal and edge sharpening simultaneously, which can play an important role in remote sensing field.

Feedback ghost imaging by gradually distinguishing and concentrating onto the edge area

Applications of ghost imaging are limited by the requirement on a large number of samplings. Based on the observation that the edge area contains more information thus requiring a larger number of samplings, we propose a feedback ghost imaging strategy to reduce the number of required samplings. The field of view is gradually concentrated onto the edge area,with the size of illumination speckles getting smaller. Experimentally, images of high quality and resolution are successfully reconstructed with much fewer samplings and linear algorithm.

Bimorph deformable mirror based adaptive optics scanning laser ophthalmoscope for retina imaging in vivo

A bimorph deformable mirror （DM） with a large stroke of more than 30 μm using 35 actuators is presented and characterized for an adaptive optics （AO） confocal scanning laser ophthalmoscope application. Facilitated with a Shack-Hartmann wavefront sensor, the bimorph DM-based AO operates closed-loop AO corrections for hu- man eyes and reduces wavefront aberrations in most eyes to below 0.1 μm rms. Results from living eyes, including one exhibiting ~5D of myopia and ~2D of astigmatism along with notable high-order aberrations, reveal a prac- tical efficient aberration correction and demonstrate a great benefit for retina imaging, including improving resolution, increasing brightness, and enhancing the contrast of images.

Nature of complex number and complex-valued neural networks

We discuss the nature of complex number and its effect on complex-valued neural networks(CVNNs).After we review some examples of CVNN applications,we look back at the mathematical history to elucidate the features of complex number,in particular to confirm the importance of the phaseand-amplitude viewpoint for designing and constructing CVNNs to enhance the features.This viewpoint is essential in general to deal with waves such as electromagnetic wave and lightwave.Then,we point out that,although we represent a complex number as an ordered pair of real numbers for example,we can reduce ineffective degree of freedom in learning or self-organization in CVNNs to achieve better generalization characteristics.This merit is significantly useful not only for waverelated signal processing but also for general processing with frequency-domain treatment through Fourier transform.