Hologram classification of occluded and deformable objects with speckle noise contamination by deep learning.

Advancements in optical, computing, and electronic technologies have enabled holograms of physical three-dimensional (3D) objects to be captured. The hologram can be displayed with a spatial light modulator to reconstruct a visible image. Although holography is an ideal solution for recording 3D images, a hologram comprises high-frequency fringe patterns that are almost impossible to recognize with traditional computer vision methods. Recently, it has been shown that holograms can be classified with deep learning based on convolution neural networks. However, the method can only achieve a high success classification rate if the image represented in the hologram is without speckle noise and occlusion. Minor occlusion of the image generally leads to a substantial drop in the success rate. This paper proposes a method known as ensemble deep-learning invariant occluded hologram classification to overcome this problem. The proposed new method attains over 95% accuracy in the classification of holograms of partially occluded handwritten numbers contaminated with speckle noise. To achieve the performance, a new augmentation scheme and a new enhanced ensemble structure are necessary. The new augmentation process includes occluded objects and simulates the worst-case scenario of speckle noise.

Optimal sampled phase-only hologram (OSPOH).

A sampled phase-only hologram (SPOH) is the phase component of the hologram of an object image with pixels being sampled with a periodic grid-cross pattern. The reconstructed image of a SPOH is a sparse image with abundant empty voids and degradation in sharpness and contrast. In this paper we proposed a method based on a new sampling scheme, together with stochastic binary search (SBS), to obtain an optimal sampling lattice that can be applied to generate phase-only holograms with enhanced reconstructed image. Experimental results show that with our proposed method, the fidelity and quality of the reconstructed image are increased.

Ensemble convolutional neural network for classifying holograms of deformable objects.

Recently, a method known as "ensemble deep learning invariant hologram classification" (EDL-IHC) for classifying of holograms of deformable objects with deep learning network (DLN) has been demonstrated. However DL-IHC requires substantial computational resources to attain near perfect success rate (≥99%). In practice, it is always desirable to have higher success rate with a low complexity DLN. In this paper we propose a low complexity DLN known as "ensemble deep learning invariant hologram classification" (EDL-IHC). In comparison with DL-IHC, our proposed hologram classifier has promoted the success rate by 2.86% in the classification of holograms of handwritten numerals.

Generation of patterned-phase-only holograms (PPOHs).

A fast and non-iterative method for generating a phase-only hologram, hereafter referred to as the patterned-phase-only hologram (PPOH), is reported in this paper. Briefly, a phase mask with a periodic phase pattern is added to the source image, and converted into a hologram. Subsequently, only the phase component is retained as a phase-only hologram. Experimental evaluation reveals that the visual quality of the reconstructed images of the PPOH generated with our proposed method is favorable, and superior to that obtained with existing methods.

Generation of complementary sampled phase-only holograms.

If an image is uniformly down-sampled into a sparse form and converted into a hologram, the phase component alone will be adequate to reconstruct the image. However, the appearance of the reconstructed image is degraded with numerous empty holes. In this paper, we present a low complexity and non-iterative solution to this problem. Briefly, two phase-only holograms are generated for an image, each based on a different down-sampling lattice. Subsequently, the holograms are displayed alternately at high frame rate. The reconstructed images of the 2 holograms will appear to be a single, densely sampled image with enhance visual quality.

Nonlinearity compensation and complex-to-phase conversion of complex incoherent digital holograms for optical reconstruction.

Incoherent digital holography (IDH) can be realized by optical scanning holography or self-interference incoherent holography. Although IDH can exhibit high quality reconstruction due to its inherently speckle-free property, direct display of an incoherent hologram is a challenge because of its amplitude nonlinearity and the demand of complex modulation. In this paper we propose to compensate the amplitude nonlinearity at the object plane, and use bidirectional error-diffusion method to convert the complex-type incoherent Fresnel hologram to a phase-only Fresnel hologram for display. A spatial light modulator is used to reconstruct the phase-only hologram optically to demonstrate the validity of our proposed method.

Fast generation of digital holograms based on warping of the wavefront recording plane.

This paper reports a fast method for generating a 2048x2048 digital Fresnel hologram at a rate of over 100 frames per second. Briefly, the object wave of an image is nonuniformally sampled and generated on a wavefront recording plane (WPR) that is close to the object scene. The sampling interval at each point on the WRP image is then modulated according to the depth map. Subsequently, the WRP image is converted into a hologram. The hologram generated with our proposed method, which is referred to as the warped WRP (WWRP) hologram, is capable of presenting a 3-D object with faster speed as compared with existing methods.

Fast conversion of digital Fresnel hologram to phase-only hologram based on localized error diffusion and redistribution.

Past research has demonstrated that a digital, complex Fresnel hologram can be converted into a phase-only hologram with the use of the bi-direction error diffusion (BERD) algorithm. However, the recursive nature error diffusion process is lengthy and increases monotonically with hologram size. In this paper, we propose a method to overcome this problem. Briefly, each row of a hologram is partitioned into short non-overlapping segments, and a localized error diffusion algorithm is applied to convert the pixels in each segment into phase only values. Subsequently, the error signal is redistributed with low-pass filtering. As the operation on each segment is independent of others, the conversion process can be conducted at high speed with the graphic processing unit. The hologram obtained with the proposed method, known as the Localized Error Diffusion and Redistribution (LERDR) hologram, is over two orders of magnitude faster than that obtained by BERD for a 2048×2048 hologram, exceeding the capability of generating quality phase-only holograms in video rate.

Generation of phase-only Fresnel hologram based on down-sampling.

We present a novel non-iterative method for generating phase-only Fresnel holograms. The intensity image of the source object scene is first down-sampled with uniform grid-cross lattices. A Fresnel hologram is then generated from the intensity and the depth information of the sampled object points. Subsequently, only the phase component of the hologram is preserved, resulting in a pure phase hologram that we call the sampled-phase-only hologram (SPOH). Experimental evaluation reveals that the numerical, as well as the optical reconstructed images of the proposed phase-only hologram derived with our method are of high visual quality. Moreover, the reconstructed optical image is brighter, and less affected by phase noise contamination on the hologram as compared with those generated with existing error-diffusion approaches.

Review of holographic-based three-dimensional object recognition techniques [invited].

With the advancement of computing and optical technologies, it is now possible to capture digital holograms of real-life object scenes. Theoretically, through the analysis of a hologram, the three-dimensional (3D) objects coded on the hologram can be identified. However, being different from an optical image, a hologram is composed of complicated fringes that cannot be analyzed easily with traditional computer vision methods. Over the years, numerous important research investigations have been attempted to provide viable solutions to this problem. The aim of this work is three-fold. First, we provide a quick walkthrough on the overall development of holographic-based 3D object recognition (H3DOR) in the past five decades, from film-based approaches to digital-based innovation. Second, we describe in more detail a number of selected H3DOR methods that are introduced at different timelines, starting from the late sixties and then from the seventies, where viable digital holographic-based 3D recognition methods began to emerge. Finally, we present our work on digital holographic, pose-invariant 3D object recognition that is based on a recently introduced virtual diffraction plane framework. As our method has not been reported elsewhere, we have included some experimental results to demonstrate the feasibility of the approach.