For the features of the space-based starry images, this paper analyzed the reason of algorithm timeout and high false alarm when detecting the orbits of moving objects under the starry background. The heavy computing of stellar identification is avoided by introducing and improving the Moving Target Indicator (MTI) filter to process the sequence starry-images. At the same time, we changed the algorithm structure from DBT to TBD and largely improved the algorithm speed under high precision. The simulation shows that the improved algorithm has the character of low time consumption, and we can use it to control the algorithm timeout and high false alarm, which meets the project needs for space-based starry-images moving objects orbits detecting.

In order to simplify the process and improve the accuracy of ship-borne star sensor install matrix calibration, a dynamic method which uses star sensor information and INS information to calculate install matrix is proposed. Theproposed method divides the calibration process into coarse calibration step and fine calibration step. Coarse calibration step calculates the coarse install matrix without consider the influence of atmospheric refraction. By using coarse install matrix and the atmospheric refraction compensation algorithm model, the atmospheric refraction is compensated and the accurate install matrix is calibrated in fine calibration step. Experiment results show that the proposed method can get the accurate calibration result by only two iterations and the install matrixes calibrated in different times are consistent. The contrastive results of star sensor attitude and INS attitude show the correctness of the proposed method.

Point Spread Function (PSF) is an important parameter indicating performance of camera and the important basis for image restoration. In order to solve the problem of uncertainty of initial phase and lack of sample points with single knife-edge, a new method named multi-phase-knife is proposed including design and data processing. Simulation result of two PSF measurement methods based on Monte-Carlo method show that relative measurement error of this method is steady and fewer than 2% by contrast, and relative measurement error of single knife-edge is only about 10%. This method is used in PSF measurement of GF-2 pan-chromatic camera, and the result is validated by double-point, which provides references of engineering application for TDI-CCD PSF measurement.

The Locality Preserving Projection (LPP) algorithms have been extensively applied for feature extraction of high dimensional face images, but selecting the neighborhood size and defining the affinity weight have a significant impact on the efficiency of LPP algorithms. In this paper, a new sparse manifold learning method was proposed, called Sparse Similarity Preserving (SSP), for dimensionality reduction of face images. It adaptively selected the similarity relation in the global structure of the data and constructed non-negative sparse graph using the sparse coefficients which reserved the global sparsity and non-linear manifold structure of face images, effectively extracting the low dimensional discriminant features. Experiments on two popular face databases (Extended Yale B, and CMU PIE), whose recognition rate reached 87.35% and 90.09%, demonstrated the effectiveness of the presented SSP algorithm.

For the high computational complexity of sparse description which results in the difficulty in meeting the actual needs and the number of samples sensitive training, we proposed a differences Shearlet characteristic of fast sparse description method to describe face recognition. First, Shearlet was used to get multi-scale and multi-direction facial image. Then a matching score fusion strategy was used to integrate Shearlet characteristics, and discriminative characteristics were constituted. Furthermore, configured the "best" sparse description for each test sample and calculated the correlated coefficient. Finally, according to the contribution size of the training sample in a test sample description, achieved the test sample image classification. Experimental results on ORL and YALE face database show that the algorithm ensuring a high recognition rate advantage as well as significantly reducing the time complexity.

Under the influence of factors such as the variations in illumination, facial expressions and noises, the accuracy of many current face recognition methods is not very satisfactory. A face recognition method named FW-SRC isproposed which is based on the fusion of wavelet packet sub-images (FW) and Sparse Representation-based Classification (SRC) via the combining research and analysis of Principal Component Analysis (PCA), wavelet packet decomposition (WPD) of images and SRC. The proposed method extracted features from the weighted fusion images of wavelet packet sub-images and then constructed the feature space by the weighted fusion images, and constructed sparse dictionary by projection of samples on the feature space, at last classified the faces by SRC. Experimental results on the Yale B, AR and CMU PIE face databases for the robustness test show that the proposed method improves the accuracy for face recognition and it has a good performance.

A new method of face recognition, which is based on phase measuring profilometry, is proposed to solve the problems of the ambient light influence in 2-D face recognition and the dependence on calibration data in 3-D face recognition. Firstly, the phase information and reflectivity information of face are obtained based on phase measuring profilometry. The phase image contains 3-D information and the reflectivity image is immune to the ambient light. Then the traditional 2-D face recognition approaches can be used after normalization. In experiments, principal component analysis and nearest neighbor method are used to classify the face. Experiment results show the effectiveness of the proposed method, which achieves higher recognition accuracy than that achieved by using grayscale image and range image.

Eyelash detection is an important step during the period of iris recognition preprocessing. An eyelash extraction algorithm based on improved ant colony algorithm is proposed. Firstly, by introducing direction factor of inside the eyelash region and outside the eyelash region, the artificial ant can gather to the eyelash-edge quickly, increasing pheromone intensity of eyelash-edge and decreasing that of non-eyelash-edge. At the same time, global and local strategies are taken respectively to update pheromone. After that, the OTSU algorithm is used to segment eyelash edge according to the pheromone intensity of image. Finally, by integrating the eyelash-edge and deleting the noise, the eyelash can be obtained. Experimental results show that, compared with other concerned algorithms, the effect of the proposed method is better than that of others, and when the distance between artificial ants is about 25 pixels, the algorithm can not only ensure eyelash detection speed but guarantee the detection effect.

A novel dynamic gesture segmentation method is proposed by combining three-frame difference method and skin-color elliptic boundary model. Firstly, the possible dynamic gesture region is determined by extracting the target property of movement via three-frame difference method, and then the target area is gotten by skin-color detection in thepossible region using skin-color elliptic boundary model. By extracting double features of dynamic hand gestures, the proposed method can effectively solve the problem that the large area of skin-color background was mistakenly identified as hands. The area of face and bare arms are removed by setting a reasonable threshold. Furthermore, for overcoming the shortcoming of traditional three-frame difference method that cannot detect static gestures, the proposed method has been improved and shown robustness on relative static gestures. The experimental results demonstrate the efficiency for dynamic hand gestures segmentation and the proposed method is suitable for real-time systems such as dynamic gesture recognition.

A modular laser 3-dimensional scanning system based on the rotary-laser theodolite positioning network is designed. As a positioning measurement system combining the measuring principle of frontage intersection of theodolite and the concept of GPS, the Rotary-laser Theodolite Positioning System (RLTPS) can realize the parallel measurement of spatial coordinate of multiple Photodiode Detector (PD). The spatial position of the laser probe is determined by the coordinates of the three PD, therefore, the portable laser 3-dimensional scanning system is realized, which can be used to the measurement in the industrial fields. The feature point cloud data of objects surface are measured by the designed Modular Laser 3-dimensional Scanning System and the experimental results show that the measurement system can satisfy the application requirements.

As there is error in laser target detecting impact points in airborne artillery air-to-ground attack training, a data processing method is proposed which eliminates singular points and averages for the single impact point in single fire, and replaces the measured points with regression fitting points for multiple impact points in continuous fire. This paper tests its effectiveness, which can provide practical significance to reduce error and evaluate training results objectively.

In order to measure retinal oxygen saturation, a dual-wavelength retinal oximeter based on commercial fundus camera was designed. The system made the retinal image of the commercial fundus camera second imaging, and simultaneously collected dual-wavelength images that respectively represented the same absorption coefficient wavelength images and the different one of oxygenated hemoglobin and de-oxygenated hemoglobin. The retinal oxygen saturation was finally measured after image registration, vessel segmentation and calculation of optic density ratio of the two images. The means oxygen saturation of artery and venules in health human for repeated measurements were 92.85% and 56.75%, respectively. The mean standard deviation was 2.41% for arterioles and 3.63% for venules.The data demonstrate that the system is reliable and reproducible. The system can obtain retinal structural imaging, which provides a useful tool for research of life sciences and diagnosis of retinal-related diseases.

With the development of remote sensing technology, high-resolution urban remote sensing images contain more structural and textural information of buildings. Buildings in urban mostly have clear corners and homogeneous roofregions. However, the instability of imaging conditions usually causes blur, light changes and other affine transformations to remote sensing images. Combined with Hessian-Affine and MSER, a fused affine region detection algorithm is proposed. Regions highly covered by others are selected according to the overlap error. Then these selected regions are considered whether to be deleted according to the affine match score. The building images’ average repeatability and number of correspondence are used for evaluation and analysis on detection. Experiments results show that the proposed method make full use of the two complementary detectors, and it obtains the best average repeatability, less redundancy under the different types of transformations. Therefore, the proposed method is better for urban remote sensing application fields.

Highly accurate displacement estimation method is critical for Ultrasound Elastography. In this study, a novel sub-sample displacement estimation algorithm where axial and lateral motion estimates are simultaneously performed to enhance the accuracy of motion tracking the proposed method has been implemented in CUDA. Compared to the original C Mex (MATLAB VERSION 2014B, Mathswork Inc., MA, USA) implementation, the GPU implementation showed a factor of 70+ acceleration while maintaining the excellent motion tracking accuracy.

According to light-emitting diode (LED) light source’s characteristic and the principles of non-imaging optics, a simple computation method is proposed which can control light path through adjusting the direction component of free-form surface. Some reflectors suitable for different application were designed by this algorithm which based on a single LED light source. Then optical entity was imported into TracePro software for non-sequential ray tracing. As a result of simulation, the optical system achieves uniform illumination. This algorithm brings a new idea for non-imaging design. It’s suitable to design microscope lighting and projection display that require high uniform lighting system.