Wolfgang Osten

A method of object recognition for single pixel imaging

Computational ghost imaging(CGI), utilizing a single-pixel detector, has been extensively used in many fields. However, in order to achieve a high-quality reconstructed image, a large number of iterations are needed, which limits the flexibility of using CGI in practical situations, especially in the field of object recognition. In this paper, we purpose a method utilizing the feature matching to identify the number objects. In the given system, approximately 90% of accuracy of recognition rates can be achieved, which provides a new idea for the application of single pixel imaging in the field of object recognition

Single shot imaging through turbid medium and around corner using coherent light

Optical imaging through turbid media and around corner is a difficult challenge. Even a very thin layer of a turbid media, which randomly scatters the probe light, can appear opaque and hide any objects behind it. Despite many recent advances, no current method can image the object behind turbid media with single record using coherent laser illumination. Here we report a method that allows non-invasive single-shot optical imaging through turbid media and around corner via speckle correlation. Instead of being as an obstacle in forming diffractionlimited images, speckle actually can be a carrier that encodes sufficient information to imaging through visually opaque layers. Optical imaging through turbid media and around corner is experimentally demonstrated using traditional imaging system with the aid of iterative phase retrieval algorithm. Our method require neither scan of illumination nor two-arm interferometry or long-time exposure in acquisition, which has new implications in optical sensing through common obscurants such as fog, smoke and haze.

Moving target detection in flash mode against stroboscopic mode by active range-gated laser imaging

Moving target detection is important for the application of target tracking and remote surveillance in active range-gated laser imaging. This technique has two operation modes based on the difference of the number of pulses per frame: stroboscopic mode with the accumulation of multiple laser pulses per frame and flash mode with a single shot of laser pulse per frame. In this paper, we have established a range-gated laser imaging system. In the system, two types of lasers with different frequency were chosen for the two modes. Electric fan and horizontal sliding track were selected as the moving targets to compare the moving blurring between two modes. Consequently, the system working in flash mode shows more excellent performance in motion blurring against stroboscopic mode. Furthermore, based on experiments and theoretical analysis, we presented the higher signal-to-noise ratio of image acquired by stroboscopic mode than flash mode in indoor and underwater environment.

Multiple-stage pure phase encoding with biometric information

In recent years, many optical systems have been developed for securing information, and optical encryption/encoding has attracted more and more attention due to the marked advantages, such as parallel processing and multiple-dimensional characteristics. In this paper, an optical security method is presented based on pure phase encoding with biometric information. Biometric information (such as fingerprint) is employed as security keys rather than plaintext used in conventional optical security systems, and multiple-stage phase-encoding-based optical systems are designed for generating several phase-only masks with biometric information. Subsequently, the extracted phase-only masks are further used in an optical setup for encoding an input image (i.e., plaintext). Numerical simulations are conducted to illustrate the validity, and the results demonstrate that high flexibility and high security can be achieved.

Fusion of visible and near-infrared images based on luminance estimation by weighted luminance algorithm

In a low-light scene, capturing color images needs to be at a high-gain setting or a long-exposure setting to avoid a visible flash. However, such these setting will lead to color images with serious noise or motion blur. Several methods have been proposed to improve a noise-color image through an invisible near infrared flash image. A novel method is that the luminance component and the chroma component of the improved color image are estimated from different image sources [1]. The luminance component is estimated mainly from the NIR image via a spectral estimation, and the chroma component is estimated from the noise-color image by denoising. However, it is challenging to estimate the luminance component. This novel method to estimate the luminance component needs to generate the learning data pairs, and the processes and algorithm are complex. It is difficult to achieve practical application. In order to reduce the complexity of the luminance estimation, an improved luminance estimation algorithm is presented in this paper, which is to weight the NIR image and the denoised-color image and the weighted coefficients are based on the mean value and standard deviation of both images. Experimental results show that the same fusion effect at aspect of color fidelity and texture quality is achieved, compared the proposed method with the novel method, however, the algorithm is more simple and practical.

The implementation of depth measurement and related algorithms based on binocular vision in embedded AM5728

Depth measurement is the most basic measurement in various machine vision, such as automatic driving, unmanned aerial vehicle (UAV), robot and so on. And it has a wide range of use. With the development of image processing technology and the improvement of hardware miniaturization and processing speed, real-time depth measurement using dual cameras has become a reality. In this paper, an embedded AM5728 and the ordinary low-cost dual camera is used as the hardware platform. The related algorithms of dual camera calibration, image matching and depth calculation have been studied and implemented on the hardware platform, and hardware design and the rationality of the related algorithms of the system are tested. The experimental results show that the system can realize simultaneous acquisition of binocular images, switching of left and right video sources, display of depth image and depth range. For images with a resolution of 640 × 480, the processing speed of the system can be up to 25 fps. The experimental results show that the optimal measurement range of the system is from 0.5 to 1.5 meter, and the relative error of the distance measurement is less than 5%. Compared with the PC, ARM11 and DMCU hardware platforms, the embedded AM5728 hardware is good at meeting real-time depth measurement requirements in ensuring the image resolution.

Optoelectronic imaging of speckle using image processing method

A detailed image processing of laser speckle interferometry is proposed as an example for the course of postgraduate student. Several image processing methods were used together for dealing with optoelectronic imaging system, such as the partial differential equations (PDEs) are used to reduce the effect of noise, the thresholding segmentation also based on heat equation with PDEs, the central line is extracted based on image skeleton, and the branch is removed automatically, the phase level is calculated by spline interpolation method, and the fringe phase can be unwrapped. Finally, the imaging processing method was used to automatically measure the bubble in rubber with negative pressure which could be used in the tire detection.

3D super resolution range-gated imaging for canopy reconstruction and measurement

In this paper, we proposed a method of canopy reconstruction and measurement based on 3D super resolution range-gated imaging. In this method, high resolution 2D intensity images are grasped by active gate imaging, and 3D images of canopy are reconstructed by triangular-range-intensity correlation algorithm at the same time. A range-gated laser imaging system(RGLIS) is established based on 808 nm diode laser and gated intensified charge-coupled device (ICCD) camera with 1392´1040 pixels. The proof experiments have been performed for potted plants located 75m away and trees located 165m away. The experiments show it that can acquire more than 1 million points per frame, and 3D imaging has the spatial resolution about 0.3mm at the distance of 75m and the distance accuracy about 10 cm. This research is beneficial for high speed acquisition of canopy structure and non-destructive canopy measurement.

Implementation and image processing of a multi-focusing bionic compound eye

In this paper, a new BCE with multi-focusing microlens array (MLA) was proposed. The BCE consist of detachable micro-hole array (MHA), multi-focusing MLA and spherical substrate, thus allowing it to have a large FOV without crosstalk and stray light. The MHA was fabricated by the precision machining and the parameters of the microlens varied depend on the aperture of micro-hole, through which the implementation of the multi-focusing MLA was realized under the negative pressure. Without the pattern transfer and substrate reshaping, the whole fabrication method was capable of accomplishing within several minutes by using microinjection technology. Furthermore, the method is cost-effective and easy for operation, thus providing a feasible method for the mass production of the BCE. The corresponding image processing was used to realize the image stitching for the sub-image of each single microlens, which offering an integral image in large FOV. The image stitching was implemented through the overlap between the adjacent sub-images and the feature points between the adjacent sub-images were captured by the Harris point detection. By using the adaptive non-maximal suppression, numerous potential mismatching points were eliminated and the algorithm efficiency was proved effectively. Following this, the random sample consensus (RANSAC) was used for feature points matching, by which the relation of projection transformation of the image is obtained. The implementation of the accurate image matching was then realized after the smooth transition by weighted average method. Experimental results indicate that the image-stitching algorithm can be applied for the curved BCE in large field.

Hyperspectral light field image denoising

Hyperspectral light field (HSLF) images with enriched spectral and angular information provide better representation of real scenes than conventional 2D images. In this paper, we propose a novel denoising method for HSLF images. The proposed method consists of two main steps. First, we generalize the intrinsic tensor sparsity (ITS) measure previously used for 3D hyperspectral image denoising to the 5D HSLF, by using the global correlation along the spectral dimension and the nonlocal similarity across the spatial and angular dimensions. Second, we further exploit the spatial-angular correlation by integrating light field super-resolution (SR) into the denoising process. In this way, the 5D HSLF can be better recovered. Experimental results validate the superior performance of the proposed method in terms of both objective and subjective quality on a self-collected HSLF dataset, in comparison with directly applying the state-of-the-art denoising methods.