Regional Time-Series Coding Network and Multi-View Image Generation Network for Short-Time Gait Recognition.

Gait recognition is one of the important research directions of biometric authentication technology. However, in practical applications, the original gait data is often short, and a long and complete gait video is required for successful recognition. Also, the gait images from different views have a great influence on the recognition effect. To address the above problems, we designed a gait data generation network for expanding the cross-view image data required for gait recognition, which provides sufficient data input for feature extraction branching with gait silhouette as the criterion. In addition, we propose a gait motion feature extraction network based on regional time-series coding. By independently time-series coding the joint motion data within different regions of the body, and then combining the time-series data features of each region with secondary coding, we obtain the unique motion relationships between regions of the body. Finally, bilinear matrix decomposition pooling is used to fuse spatial silhouette features and motion time-series features to obtain complete gait recognition under shorter time-length video input. We use the OUMVLP-Pose and CASIA-B datasets to validate the silhouette image branching and motion time-series branching, respectively, and employ evaluation metrics such as IS entropy value and Rank-1 accuracy to demonstrate the effectiveness of our design network. Finally, we also collect gait-motion data in the real world and test them in a complete two-branch fusion network. The experimental results show that the network we designed can effectively extract the time-series features of human motion and achieve the expansion of multi-view gait data. The real-world tests also prove that our designed method has good results and feasibility in the problem of gait recognition with short-time video as input data.

Multi-wavelength spatial frequency domain diffuse optical tomography using single-pixel imaging based on lock-in photon counting.

We present a spatial frequency domain (SFD) diffuse optical tomography for simultaneous acquisition of multi-wavelength tomographic images of turbid media. We propose a highly sensitive single-pixel SFD imaging system for simultaneously collecting multi-wavelength spatially modulated reflectance images, instead of using the expensive electron-multiplying charge-coupled device camera that requires switching between the multi-wavelength collections. The single-pixel SFD imaging system using three low-power light sources (455, 532 and 660 nm) that were intensity-modulated by square waves with three different frequencies for frequency encoding, and all the light sources were focused onto one digital micromirror device (DMD) for generating wide-field sinusoidal illumination patterns. Reflected light from the surface of the turbid media was modulated by the other DMD with many sampling patterns before being spatially integrated. Spatially integrated light signals were frequency decoded with a novel highly sensitive lock-in photon counting detection, then multi-wavelength spatially modulated reflectance images were recovered with the single-pixel imaging (SPI) method. We incorporated the two-dimensional discrete cosine transform (DCT) into the SPI method to reduce the number of sampling patterns, and, thereby, the proposed DCT-SPI scheme achieved a fast acquisition of SFD reflectance images that is desired for a dynamic SFD imaging application. Direct current (DC) and alternating current (AC) amplitudes at all the locations on the media surface were extracted from the recovered images. Multi-wavelength tomographic images were reconstructed with an inversion algorithm based on the first-order Rytov approximation of the diffusion equation, using both the extracted DC and AC amplitudes. We performed experiments using a series of tissue simulating phantoms to verify the performances of the proposed approach and compared the experimental results with those using a conventional camera-based SFD imaging system. The results demonstrate that our DCT-SPI based SFD-DOT approach is well suited for simultaneous reconstruction of multi-wavelength tomographic images to pave the way for many SFD imaging applications.

Determination of optical properties of turbid medium from relative interstitial CW radiance measurements using the incomplete P3 approximation.

Interstitial determination of the tissue optical properties is important in biomedicine, especially for interstitial laser therapies. Continuous wave (CW) radiance techniques which examine light from multiple directions have been proposed as minimally invasive methods for determining the optical properties under an interstitial probe arrangement. However, both the fitting algorithm based on the P3 approximation and the analytical method based on the diffusion approximation (DA), which are currently used recovery algorithms, cannot extract the optical properties of tissue with low transport albedos accurately from radiance measurements. In this paper, we proposed an incomplete P3 approximation for the radiance, the P3in for short, which is the asymptotic part of the solution for the P3 approximation. The relative differences between the P3in and the P3 were within 0.48% over a wide range of clinically relevant optical properties for measurements at source detector separations (SDS) from 5 mm to 10 mm and angles from 0° to 160°. Based on the P3in, we developed an analytical method for extracting the optical properties directly using simple expressions constructed from the radiance measurements at only two SDSs and four angles. The developed recovery algorithm was verified by simulated and experimental radiance data. The results show that both the absorption and reduced scattering coefficients were recovered accurately with relative errors within 5.28% and 3.86%, respectively, from the simulated data and with relative errors within 10.82% and 10.67%, respectively, from the experimental data over a wide range of albedos from 0.5 to 0.99. Since the developed P3in-based radiance technique can obtain the optical properties rapidly from the measurements at only two SDSs and four angles, it is expected to be used for in vivo and in situ determination of the optical properties in online treatment planning during laser therapies.

Wide-field fluorescence tomography with composited epi-illumination of multi-frequency sinusoidal patterns.

We present a spatial-frequency domain (SFD) fluorescence tomography (FT) for acquiring three-dimensional fluorophore distribution in turbid media. The approach uses a composited epi-illumination of multi-frequency sinusoidal patterns on a sample of semi-infinite geometry and demodulates the measured data with a generalized phase shifting scheme to calculate the modulation transfer function (MTF) at each frequency. This method results in a significantly reduced number of the optical field measurements, as compared to those with separate illumination of single-frequency sinusoidal patterns, and, thereby, achieves a fast data acquisition that is desired for a dynamic imaging application. Fluorescence yield images are recovered with the normalized Born formulated inversion of the diffusion model by simultaneously using the multi-frequency MTFs. Simulative and experimental reconstructions are performed in comparison with the single-frequency scheme to validate the proposed algorithm. The results suggest that adoption of the multi-frequency strategy to the SFD-FT can substantially improve the reconstruction quality, as well as its imaging resolution and quantitative accuracy.

An endoscopic diffuse optical tomographic method based on the effective detection range.

Endoscopic diffuse optical tomography (DOT) is a new medical imaging modality with the potential applications in functional imaging of the internal organs. To cut down the measurement time and the computation burden of image reconstruction, in this paper, we developed the image reconstruction algorithm with the partial measurement in the effective detection range (EDR) of a tubular tissue and the corresponding endoscopic imaging system with a novel endoscopic probe for flexibly selecting the detection sites. For a typical inner size and optical properties of the cervix, it is found that EDR is less than half of the inner circumference. Comparing to the traditional method, the adoption of EDR results in a reduction of more than a factor of two in the time cost for a measurement cycle and for the total iteration reconstruction. Images reconstructed from the simulation data demonstrate that the proposed method achieves equivalent image quality to that obtained from the complete data set. The images reconstructed from the EDR measurements on cervix-like solid phantoms show that both the location and size of the targets are reconstructed correctly. The proposed method will be useful to the development of endoscopic DOT technologies for cancer detection in tubular organs including cervix.