In order to improve resolution of single frame image with severe degradation, we propose a novel super-resolution reconstruction framework via regularization restoration combined with learning-based sparse representation enhancement. To achieve enlargement and suppression of blurring and noise simultaneously, we carefully balance the data fidelity and the prior item using regularization parameter on the basis of verisimilar estimation of degradation. Based on the acquired relatively clean image and pre-constructed over-complete sparse representation dictionary, image resolution zooming with characteristic of edge-preserving can then be realized. Fundamentally, the output of preceding regularization reversion remarkably betters low-frequency quality of bicubic interpolation version in conventional learning-based super-resolution. Furthermore, the ridding of blur and noise can favorably weaken dependency of atoms to degraded information. Consequently, their combination of two techniques can remarkably eliminate blur and noise, and meanwhile, remove annoying edge artifacts of enlarged image. Experiment results demonstrate that the addressed approach produces visually pleasing magnification for blurry and noisy low-resolution image.

A novel feature extraction algorithm of ultrasound image combined with the texture, shape and the attenuation characteristics information was proposed, which could be used to identify the benign or malignant thyroid tumors. This paper focused on improving the texture feature extraction algorithm of thyroid tumors. On the basis of the traditional Local Binary Pattern (LBP) algorithm, we extended the neighborhood distribution in the elongated manner, which was more conducive to describe thyroid tumors and extract anisotropic properties of tumor effectively. Also, we used fuzzy logic to encode distance, which overcame the uncertainty of speckle noise in the ultrasound image. Furthermore, we extracted tumor circularity, standard deviation of the normalized radial length, area ratio, roughness index and the attenuation coefficient, which formed a feature vector to characterize thyroid tumors. Finally, Support Vector Machine (SVM) was used to classify and identify the thyroid nodules. Compared with other methods of feature extraction, the proposed feature fusion algorithm has a high accuracy of description, which can achieve higher classification accuracy, and the reasonableness and effectiveness of the proposed method is verified by experiments.

Denoising the high resolution remote sensing images is a difficult problem in the relative research field of remote sensing. A novel algorithm for denoising the high resolution remote sensing images is proposed based on sparse representation. A dictionary which has an efficient description of remote sensing image content is obtained based on K-SVD algorithm according to the characteristics of the added noise of high spatial resolution remote sensing images. Denoising is realized by using sparse representation, and the useful information of the image is kept. The experimental results of the remote sensing images obtained by “the first satellite of high resolution” show that the algorithm can filter out the noise in the image more effectively and improve the PSNR, and this method has better performance than other dictionary learning algorithms and other denoising algorithms.

A new method of motion estimation based on the search of the local extremum was proposed to reduce the time of motion estimation in the video encoding. Firstly, divided the search area into several parts according to the size of the search blocks; Secondly, started to find the local extremum of the search part in spiral way and compared it with the extremum ahead. In this step, early termination strategy and the algorithm of marking and skipping were proposed so as to further speed up the search. In the last, took the minimum value of the search parts as the global minimum and searched that part with the diamond pattern in the step of 1, which could get the perfect location. Test results show that compared with the UMH, the new method can improve the encoding speed but the PSNR increases less than 0.03 kb/s and the bitstream rate decreases less than 0.005 dB. What’s more, this method can reduce the encoding time by 10% compared with the UMH when encoding the video which has very high motion complexity.

For Gaussian blur distorted stereoscopic images, a no-reference image quality assessment method is proposed based on the Bidimensional Empirical Mode Decomposition (BEMD). In the method, intrinsic mode function components and residual image are firstly produced by applying BEMD. Then, the statistical information (mean, variance and entropy), and shape and scale parameters are obtained by applying generalized Gaussian distribution method on the intrinsic mode function components, to form stereoscopic image feature information. Finally, support vector regression is performed to predict the objective scores by establishing the relationship between the stereoscopic image features and the subjective scores. Experimental results show that, the proposed method can achieve higher consistency with subjective assessment of stereoscopic images for Gaussian blur distorted stereoscopic images.

A multispectral imaging system acquires spectral color images by splitting the light spectrum into multiple bands. Due to the different refractive index for individual wavelength, the focus positions of the channels are different, and thus each channel should be adjusted to obtain sharp image. We propose a fast autofocus method based on the similarity of focus measure distributions, which is suitable to multispectral imaging. In the proposed method, we first obtain the focus measure distribution of a reference channel using traditional autofocus method. Then, the clear focus positions of other channels are located according to the distribution similarity between individual channels. As only a few images are needed, the autofocus for these channels is quite fast. Experimental results validate that the proposed method performs better than the traditional techniques.

Referring to a new type of quasi-reciprocal reflective optical voltage sensor optical circuit structure and according to the environment temperature changes impact on polarization fiber birefringence, a method of constructing ideal reciprocity optical circuit is worked out. In actual application, this method can quickly and accurately identify the best length of measuring optical fiber and compensating optical fiber, and make the light path achieve better reciprocity, which effectively improves the anti-interference ability of the optical circuit. At the same time, the optical circuit can obtain good temperature correlation. The practical application and the production of this kind of optical circuit have a great reference value.

For decreasing the high grating inscription scattering of previous array-hole microstructured Optical Fiber Grating (MOFG), a large circular side-hole microstructured optical fiber grating configuration is proposed and analyzed. In this new configuration, a section of array holes is replaced by one large circular hole that is filled with fluid of proper fluidic index. Simulation shows that, due to immunity of array hole and microlens effect, the grating inscription efficiency of this proposed structure is respectively 288% and 206% times of the conventional array hole microstructured optical fiber grating and Single Model Fiber Grating (SMFG), which shows its potential value in high efficient grating. Moreover, a symmetrical angle inscription method is also proposed to guarantee the symmetry of grating inscription field, which will contribute to improving the flatness of beam energy distribution and the corresponding birefringence in single angle inscription process.

We propose and fabricate a high-sensitivity Mach-Zehnder interferometer (MZI) refractive index (RI) sensor formed by concatenating two Core-offset Joints (COJs) in a cladding-etched fiber. The two COJs are fabricated in a single-mode fiber by using a commercial fiber fusion splicer. Then, the fiber cladding is thinned by chemical etching, and high-sensitivity MZI RI sensor is formed owing to the strong interaction of the evanescent field of the thin fiber and the surrounding medium. Experimental results show that, for the MZI with a length of 10 mm and a fiber diameter of 13.2 μm, its RI sensitivities are up to 1 288.05 and 2 118.26 nm/RIU in the RI ranges of 1.333~1.368 5 and 1.368 5~1.386, respectively. The sensor reveals the advantages of high sensitivity, compact size, simple fabrication, as well as low cost.

Considering the requirement of high accuracy, a novel Fiber Bragg Grating (FBG) modulation method to fit the FBG wavelength with multi-peak reflectivity is proposed. Application of Levenberg-Marquardt algorithm in fitting of peak-reflectivity by sidebands was discussed. The relation model was researched by reflecting central wavelength and peak-reflectivity of sidebands based on the reflex action of FBG for multi-wavelength source. Levenberg-Marquardt algorithm was used to extract the characters from the reflectance spectra to mark sensing information. Temperature resolution 0.1℃ and strain resolution 0.5 με is respectively obtained. The experimental results show that the new method has something referential value to give a guarantee to suit the high accurate measurement with FBG monitoring system.

Interface is studied to solve the heat dissipation problem of the LED street lamp, improve the stability of the overall performance of lamps and promote the LED luminous efficiency. Firstly we analyzed the rationality and defects for interface of LED street lamp, such as modules fixed by four positioning Screws and Zhaga consortium of the modules. Thus we summed up three characteristics which high power LED lighting module interface should possess. According to the actual work environment and installation conditions of LED street light in our country, this article illustrates another kind of Wedge interface application in the modular LED street lamps, which is of efficient heat dissipation, convenient installation, and easy upgrading. By means of theoretical analysis and numerous practical tests, the technical advantages and generalizability of wedge connection of the LED modules are verified.

Echelle grating is used extensively in spectral instruments owing to its advantages, such as the advanced intrinsic dispersion and resolving power. But while ruling the echelle grating, there may be some periodic components of spacing errors which will seriously affect the performance of grating diffraction and lead to defects of the diffraction effects. In response to this situation, the effect of grating’s line error on diffraction spectrum, diffraction order, wavefront error, the intensity of the rowland ghost line and the intensity of stray light was proposed. What’s more, a method for analyzing the grating’s line error from the aspect of the diffraction efficiency was proposed. Under certain conditions, the ruling errors for meeting every kind of requests are 1 265.8 nm, 352 nm, 37 nm, 112 nm, 3.4 nm respectively, which provides a direct and important reference for the ruling accuracy before ruling an echelle grating.

By introducing big screen positioning methods based on light reflection, the principle of visual touch is explained and the big touch-screen system is built. For large-scale trend of FPD (Flat Panel Display), touch detection method based on small-signal is studied. Row selected, background followed and screen heavy pressure removal method are proposed. It is demonstrated in the experiments that large screen touch system based on light reflection achieves full-screen drawing on a 100-inch screen without wearing any marks or sensors and there are not drawing breaking and skipping phenomena. The design of touch system also lays the foundation for large-size volumetric human-computer interaction.

The research of video tracking is developing forward wide-range and long-time object tracking. Pedestrian re-identification is the key technology of wide-range and long-time pedestrian tracking, and is foundation of follow-up behavior analysis. A pedestrian re-identification algorithm is proposed based on non-sparse multiple kernel Support Vector Machine (SVM). Firstly, we extract multilayer SIFT feature and multilayer color histogram feature of tracked pedestrians video image sequence. Then, we online fuse multilayer SIFT feature and multilayer color histogram feature to obtain pedestrian appearance models using non-sparse multiple kernel SVM. Finally, we re-identify pedestrian objects using the stored pedestrian appearance models. The method can be applied to the same pedestrian tracking across cameras in the multiple cameras video surveillance and recognition of pedestrian recurrences in the single camera video surveillance. The experiment results show that our method can rapidly train pedestrian object appearance models and achieve very high recognition rate.

To satisfy rigid performance specifications of real-time and accuracy while tracking and locating mouth region in an intelligent driver assistance system, a new method is proposed. The idea is implemented with two steps, tracking and target detection. In the tracking part, by making a diffusion of tracking range, and optimize process of update learning factor to achieve long time tracking; In detection and positioning section, fully taking the difference of each human mouth and different expressions into account, the accuracy required in the auxiliary driving system is achieved by use of Adaboost algorithm. In addition, through the introduction of compressed sensing and cascade classifier, rapidness is also guaranteed. Experimental results show that the method can reach the target while it is applied in driver assistance system. It is a strong significance of practical application.