Session Index

In this study, we proposed the method to conceal the eye-track camera. The image of the user's eye propagates via the waveguide combiner. The aberration caused by the waveguide was analysed via ray-tracing method. Furthermore, we also discuss the method to compensate for the aberration and enhance the image quality.

  Preview abstract

In this study, High-precision 3x3 polarization maintaining fiber coupler was made with fused biconical taper method. To control the pulling speed and temperature, the coupling ratio can be better than 31% for the three output ports and coupling ratio difference will be less than 3% between port 1 and port 3. In additions, the extinction ratio and excess loss can be reached to 20 dB and better than 0.7 dB.

  Preview abstract

In this method, we derived the phase of the asymmetry intensities of interference signals and found that the phase caused by the overlay is proportional to the diffraction order. A high order diffraction light was used to enhance the overlay measurement signal. To demonstrate the feasibility, we compared the frequencies of the asymmetry intensity curve under different overlay values. The frequency of the asymmetry intensity of the second order diffraction signal was twice of that of the first order diffraction signal.

  Preview abstract

We demonstrate the preparation of photopolymer glass films with good optical quality by sol-gel reaction. Samples with maximum diffraction efficiencies over 90% were produced. Using different angles when recording holograms shows that the samples can make different holographic diffraction elements for application in various systems. Finally, the high transmittance and low shrinkage were verified by spectrometer measurement and Bragg mismatch angle experiment.

  Preview abstract

In recent years, the technology of image recognition has been popular and used in many applications, such as the real-time detection of animals for species conservation. In the study, we use the YOLO(You Only Look Once) neural network to recognize the deer in photos for estimating the amount of deer living in some region. By using the images of sika deer captured from videos for training YOLO neural nework to find sika deers accurately.The results show that YOLO can detect the sika deer in photos and videos in real time, which 95% accuracy can be achieved.

  Preview abstract

Analyzing hyperspectral imaging (HSI) is a way to find out the material properties that are difficult to be observed with our naked eyes. This experiment was based on the hyperspectral sensor developed by imec. We used push-broom scanning to get the data of HSI. Our interface built by PyQt5 was able to display the camera image in real time and capture images with the one-dimensional motorized platform. After analyzing the deviation of stitched images, we concluded that the platform may have 10.3˚clockwise rotations relative to the camera.

  Preview abstract

In this study, a novel phase shift interferometer with moving grating as the phase shifter was proposed. To demonstrate the feasibility of this method, the phase retardations between the S and P polarizations of a quarter-waveplate and a half-waveplate were measured. This method has the advantages of high accuracy and time saved.

  Preview abstract

A binary encoding algorithm is presented for phase-shifting projected fringe profilometry. It does not require additional projections to identify fringe orders. The pattern used for phase extraction can be used for phase unwrapping directly. Fringes can be discerned even though the surface color or reflectivity varies with positions.

  Preview abstract

This research purpose is by using two different photo-polymerization 3D print to produce microoptic devices. After comparing the effectiveness of the devices, photo-polymerization 3D print has better surface contour and resolution.
In this research, I use Arduino to detect the path of motor and control the laser output of optical pickup head. Making use of small focus and step width from optical pickup head and stepper motor, we can get a 3D stereolithography effect. I use Phrozen Sonic Mini 8K LCD printer to produce devices. Finally, compare the effects of the both.


  Preview abstract

This study proposes DL-GSA, an optimized non-iterative algorithm that computes accurate holograms using an unsupervised learning convolutional neural network model. The DL-GSA is intended to improve the holographic display's computation speed and enable it to achieve real-time computation. According to the results, the average accuracy of the computer simulations generated using the DL-GSA is almost better than the Modified Gerchberg-Saxton algorithm (MGSA).

  Preview abstract

A roll angle displacement measurement technique with high resolution for large range measurement is presented. The proposed technique is based on the birefringence effect and phase detection of polarization interferometry with a polarization camera, and the dual beam configuration can greatly eliminate the interference of pitch angular displacement to the roll angular displacement measurement system. The experimental results demonstrate that the resolution of the system can reach 0.038 arcsec with a 10° measurement range.

  Preview abstract

We evaluate the performance of the mode recognition based on sparse coding for OAM beams affected by atmospheric turbulence in free-space optical communication. Since the sparse coding is a linear model machine learning process, the mode recognition results outperform the usual optics-based mode recognition using an inverse phase multiplication and a spatial filter.

  Preview abstract

A probe laser light is integrated into a microscope so that the defocus of the microscope can be easily monitored from the movement of the light spot at the image plane. By this way, the sample can be located at the focus plane with error within ± 2μm.

  Preview abstract

Optical phase reconstruction images generated by the homemade high-resolution wavefront sensor are carried out. The method of transverse differential interferometry is used to obtain the optical amplitude and phase information of microstructure, simultaneously. The wavefront sensor equipped with an optical microscope achieved the precise measurements of the phase-shift.

  Preview abstract

This study focuses on an automatic detection system integration and application algorithm to detect defective ping-pong balls based on fringe projection profilometry (FPP). The experimental results show that the proposed system can perform fair detection rate and accuracy under suitable defective types and environment conditions.

  Preview abstract

We developed a wafer warpage deformation controlling system consisting of a wafer cooling system, wafer surface profile, and temperature monitor system. Then, a correlation curve can be obtained between the wafer height change and the temperature change. Consequently, we could use this correlation curve to control the wafer surface height by adjusting the wafer surface temperature.

  Preview abstract

Diffractive structured light is widely used in 3D sensing in which the projection images are usually performed using the diffractive optical elements (DOEs). The design of DOEs is based on Fraunhofer diffraction, however, the off-axis distortion and attenuation of light intensity occurs in the realized images. These problems lead to the decrease in the corner recognition accuracy and the recognition speed. In this paper, we proposed a modified method to solve the problems of distortion and light intensity attenuation. We thereby designed and produced the binary phase DOEs, and verified the diffraction patterns which are consistent with the original targets.

  Preview abstract

A three-dimensional (3D) bright spot can be produced by specific scratches on a surface, which is called scratch stereogram. Here we calculated the position and direction of the scratches from a 3D model. The scratch stereogram is fast fabricated on a plate by CNC engraving.

  Preview abstract

This paper presents a method to divide the fast Fourier transform (FFT) algorithm into two parts: the outer part is conducted by our codes and the inner part is conducted by MATLAB default functions. The outer part divides a very large array into small arrays and, in the inner part, the small arrays are conducted by MATLAB’s FFT in a temporal sequence. The small array returns to the outer part to complete the FFT that they are supposed to do. We applied this algorithm to the iterative Fourier transform algorithm and achieved the calculation of DOE with 4840×4840 pixels.

  Preview abstract

With the advent of the global semiconductor 3-nm process era, the technology of non-contact measurement has become the core of current measurement development. In this study, an optical non-contact measurement system is developed for surface roughness measurement of wafers after wire-cutting. By introducing the relationship between surface roughness and the metrology of binary-image, producing the numerical analysis of regression curves, the surface roughness of the object to be measured can be quickly predicted.

  Preview abstract

In order to improve the experimental accuracy, we redesigned the manual XY-axis platform of the microscope to motorized platform. And write a program to control the monochromator and the CCD scanning. Compared HSI with spectrometer to obtain the data of calibration.

  Preview abstract

This research involved the classification algorithms and graphical user interface for hyperspectral images. The sharpness algorithm was added for image classification and the graphics processing unit (GPU) was adopted in spectral angle mapping (SAM) classification to compare the time consumption between central processing unit (CPU) and GPU. In this paper, we focused on the development of algorithms and the combination between algorithms and the GPU parallel computing. It is expected that the research can inspire innovation and achievability for pairing hyperspectral and machine learning.

  Preview abstract

In this study, spectral analysis was performed on retinal images of hydroxychloroquine retinopathy and dementia; after analyzing the spectrum, we conducted deep learning for dementia, using a total of 3256 retinal images. Among them, according to Dementia, mild cognitive impairment (Mci) and Normal, the images are divided into training set and test set and then brought into four neural network models for training.

  Preview abstract

In a depth scanning profilometer, the optical signals extracted from a short depth of filed are investigated. The local intensity modulation by projected fringe is analyzed and accumulated to rebuilt the surface profile of the inspected object.

  Preview abstract