基于多特征融合的人像HDR图像处理算法

基于深度学习的高动态范围(High Dynamic Range,HDR)图像处理算法在处理含有人像的图片时存在皮肤偏色问题。针对此问题,本文提出了一种基于多特征融合的人像HDR图像处理算法U2HDRnet。该算法由皮肤特征提取模块、三边特征提取模块、色彩重建模块3部分构成。首先,皮肤特征提取模块分离出皮肤区域的颜色和位置信息;其次,三边特征提取模块分别提取图片的局部特征、全局特征和语义特征,并与皮肤特征融合;最后,色彩重建模块对网格做空间和颜色深度上的插值。此外,本文引入改进的自注意力与卷积融合模块以提升HDR的处理效果;同时本文还制作了人像HDR数据集PortraitHDR,填补了该领域内数据集的空白。实验结果显示,U2HDRnet的PSNR达31.42 dB,SSIM达0.985,均优于目前常见的HDR算法,在获得高质量人像HDR图像的同时避免了皮肤的失真。

Distortion-Free Data Embedding Scheme for High Dynamic Range Images

Distortion-free data embedding is a technique which can assure that not only the secret data is correctly extracted but also the cover media is recovered without any distortion after secret data is extracted completely. Because of these advantages, this technique attracts the attention of many researchers. In this paper, a new distortion-free data embedding scheme for high dynamic range （HDR） images is proposed. By depending on Cartesian product, this scheme can obtain higher embedding capacity while maintaining the exactly identical cover image and stego image when using the tone mapping algorithms. In experimental results, the proposed scheme is superior to Yu et aL＇s scheme in regard to the embedding rate——an average embedding rate of 0.1355 bpp compared with Yn et aL＇s scheme （0.1270 bpp）.

Analysis of Wide-Bandgap Material OPFET UV Detectors for High Dynamic Range Imaging and Communication Applications

The ultraviolet (UV) photoresponses of Wurtzite GaN, ZnO, and 6H-SiC-based Optical Field Effect Transistor (OPFET) detectors are estimated with an in-depth analysis of the same considering the generalized model and the front-illuminated model for high resolution imaging and UV communication applications. The gate materials considered for the proposed study are gold (Au) and Indium-Tin-Oxide (ITO) for GaN, Au for SiC, and Au and silver dioxide (AgO2) for ZnO. The results indicate significant improvement in the Linear Dynamic Range (LDR) over the previously investigated GaN OPFET (buried-gate, front-illuminated and generalized) models with Au gate. The generalized model has superior dynamic range than the front-illuminated model. In terms of responsivity, all the models including buried-gate OPFET exhibit high and comparable photoresponses. Buried-gate devices on the whole, exhibit faster response than the surface gate models except in the AgO2-ZnO generalized OPFET model wherein the switching time is the lowest. The generalized model enables faster switching than the front-illuminated model. The switching times in all the cases are of the order of nanoseconds to picoseconds. The SiC generalized OPFET model shows the highest 3-dB bandwidths of 11.88 GHz, 36.2 GHz, and 364 GHz, and modest unity-gain cut-off frequencies of 4.62 GHz, 8.71 GHz, and 5.71 GHz at the optical power densities of 0.575 μW/cm2, 0.575 mW/cm2, and 0.575 W/cm2 respectively. These are in overall, the highest detection-cum-amplifi-cation bandwidths among all the investigated devices. The same device exhibits the highest LDR of 73.3 dB. The device performance is superior to most of the other existing detectors along with comparable LDR, thus, emerging as a high performance photodetector for imaging and communication applications. All the detectors show considerably high detectivities owing to the high responsivity values. The results have been analyzed by the photovoltaic and the photoconductive effects, and the series resistance effects and will aid in conducting further research. The results are in line with the experiments and the commercially available software simulations. The devices will greatly contribute towards single photon counting, high resolution imaging, and UV communication applications.

Multi-exposure fusion for high dynamic range scene

Due to the existing limited dynamic range a camera cannot reveal all the details in a high-dynamic range scene. In order to solve this problem,this paper presents a multi-exposure fusion method for getting high quality images in high dynamic range scene. First,a set of multi-exposure images is obtained by multiple exposures in a same scene and their brightness condition is analyzed. Then,multi-exposure images under the same scene are decomposed using dual-tree complex wavelet transform( DT-CWT),and their low and high frequency components are obtained. Weight maps according to the brightness condition are assigned to the low components for fusion. Maximizing the region Sum Modified-Laplacian( SML) is adopted for high-frequency components fusing. Finally,the fused image is acquired by subjecting the low and high frequency coefficients to inverse DT-CWT.Experimental results show that the proposed approach generates high quality results with uniform distributed brightness and rich details. The proposed method is efficient and robust in varies scenes.

Improvement of High Dynamic Range Capacitive Displacement Sensor by a Globalm Planarization

This study presents an improvement of high dynamic range contact-type capacitive displacement sensor by applying planarization. The sensor is called the contact-type linear encoder-like capacitive displacement sensor (CLECDiS), is a nano-meter-resolution sensor with a wide dynamic range. However, height differences due to patterned electrodes may cause a variety of problems or performance degradation. In devices of two glass wafer surfaces with patterned structures assembled face-to-face and in sliding contact, the heights of the patterns crucially affect their performance and practicality, so it should be planarized for reducing the problem. A number of techniques for planarizing glass wafer surfaces with patterned chrome electrodes were evaluated and the following three were selected as adequate: lift-off, etch-back, and chemical mechanical polishing (CMP). The fabricated samples showed that CMP provided the best planarization. CMP was successfully employed to produce CLECDiS with improved signal reliability due to reduced collisions between electrodes.

8K电视信号的HDR和色域扩展技术

探讨8K电视技术中高动态范围(High Dynamic Range,HDR)和色域扩展技术的应用。首先,介绍8K电视的定义、发展历程、应用领域和基本特性。其次,着重探讨HDR技术的概念和在8K电视中的实现方式,以及色域扩展技术的定义、原理、标准和在8K电视中的应用方法。通过实例验证HDR和色域扩展技术对图像质量的提升效果。最后,总结研究成果,强调8K电视技术在提供卓越视觉体验方面的潜力和重要性。

Novel Logarithmic Active Pixel Sensor with High Dynamic Range and High Output Swing

The logarithmic response complementary metal oxide semiconductor （CMOS） image sensor provides a wide dynamic range, but its drawback is the lack of simple fixed pattern noise（FPN） cancellation scheme. Designed is a novel logarithmic active pixel sensor（APS） with high dynamic range and high output swing. Firstly, the operation principle of mixed-model APS is introduced. The pixel can work in three operation modes by choosing the proper control signals. Then, FPN sources of logarithmic APS are analyzed, and double-sampled technique is implemented to reduce FPN. Finally, according to the simulation results, layout is designed and has passed design rule check（DRC）, electronic rule eheck（ERC） and layout versus schematic（LVS） verifications, and the post-simulation results are basically in agreement with the simulation results. Dynamic range of the new logarithmic APS can reach about 140 dB; and the output swing is about 750 inV. Results show that by using double sampled technique, most FPN is eliminated and the dynamic range is enhanced.

融合HDR技术的智能视频处理与分发系统设计

分析高动态范围(High Dynamic Range,HDR)视频处理与分发面临的挑战,提出一种融合HDR技术的智能视频处理与分发系统。该系统通过深度学习算法实现HDR视频的智能处理,包括场景分析、动态范围扩展和色彩增强等。在分发端,该系统采用自适应比特率流媒体技术。实验结果表明,该系统在视频质量提升、用户体验优化和传输效率提高方面均取得显著成效,为HDR视频的智能处理和高效分发提供了新的解决方案。

Shape recovery using high dynamic range images

An effective method for object shape recovery using HDRIs （high dynamic range images） is proposed. The radiance values of each point on the reference sphere and target object are firstly calculated, thus the set of candidate normals of each target point are found by comparing its radiance to that of each reference sphere point. In single-image shape recovery, a smoothness operation is applied to the target normals to obtain a stable and reasonable result; while in photometric stereo, radiance vectors of reference and target objects formed due to illuminations under different fight source directions are directly compared to get the most suitable target normals. Finally, the height values can be recovered from the resulting normal field. Because diffuse and specular reflection are handled in an unified framework with radiance, our approach eliminates the limitation presented in most recovery strategies, i.e., only Lambertian model can be used. The experiment results from the real and synthesized images show the performance of our approach.

基于密集连接编解码网络的HDR重建算法

针对动态场景下高动态范围(High Dynamic Range,HDR)重建会出现伪影现象的问题,提出一种基于密集连接编解码网络的HDR重建算法。该算法基于通道和空间维度的注意力机制模块,根据注意力特征图引导并提取不同曝光低动态范围图像相对应的图像特征,自动抑制由失准和过饱和引起的不期望分量,增强非参考图像中所期望的精细细节。通过密集连接编解码网络模块,在编码部分逐步提取特征图的多层次细节,并利用密集连接增强信息交互,使用解码器还原图像细节特征,最终生成少伪影、高质量的高动态范围图像。实验结果表明,该算法可以有效减弱伪影现象,得到细节丰富的高动态范围图像。

Fast Scheme for Projective Geometric Correction and Edge Blending Based on High Dynamic Range Images

With the development of graphic processing unit(GPU)power,it is now possible to implement geometric correction and edge blending functions on a single computer.However,the processing resources consumed by the geometric correction and edge blending phases still burden the system and slow down the main application considerably.A new platform independent scheme is proposed,minimizing the negative influence on performance.In this scheme,parameters for geometric correction and edge blending are firstly defined in an interactive way and recorded as a 32-bit high dynamic range(HDR) image,which is then used by high level shading language(HLSL) codes embedded in the main application as a lookup table,greatly reducing the computational complexity and enhancing flexibility.

High Dynamic Range Image Fusion Based on Wavelet

With the developpment of image fusion technology and the maturity of wavelet theory, wavelet transform with its good time-frequency characteristics stands out in the field of image fusion. On the basis of wavelet transforms theory, this article proposes a high dynamic range imaging confusion method which combines with wavelet decomposition. First, perform a wavelet multi-scale decomposition to the two registered source image; then conduct wavelet inverse transform to the decomposed images. This paper focuses on the characteristics of high frequency and low frequency domain after wavelet decomposition,using different fusion methods in each of the frequency domain, finally obtain the fused image through inverse wavelet transform image reconstruction. The simulation results and evaluation index results show that, compared with other similar methods, this method is better in retaining the original image's details information, and improves the quality of fusion image.

基于注意力和反馈机制的HDR视频重建

对基于深度学习的高动态范围(high dynamic range,HDR)重建进行研究,提出一种基于注意力和反馈机制的HDR重建方法.首先,将时间上连续、循环曝光的3张图像作为网络的输入,通过引入注意力模块生成注意力图像,对获取的特征进行自适应的加权,以优化网络的特征提取和减少鬼影现象的出现;然后,将反馈机制引入到网络中,进一步提高特征信息的利用率,优化网络在特征融合和重建方面的性能;最后,在L1损失函数的基础上,考虑色彩相似度损失函数和VGG(Visual Geometry Group)损失函数以增强重建后HDR图像的色彩表现及高频细节.实验结果表明,本方法不仅可获得更好的主观和客观重建质量,而且优于目前存在的主流算法.

基于区域像素亮度分析的HDR技术质量客观评价方法研究

在中央广播电视总台4K/8K超高清电视节目制播技术要求中,HDR(高动态范围)采用HLG标准/1000nit(GY/T315-2018)是4K/8K电视节目制作的必要选项。目前,对HDR技术参数的审核大多采用专业监视器主观评价方式,如何客观评价电视节目HDR技术质量是一个空白。本文根据HLG映射函数原理,提出了通过图像区域像素亮度分析的方法,对HDR技术质量进行客观分析。

基于环境光的HDR影像视觉舒适度研究

随着高动态范围(HDR)显示技术的发展,人们越来越关注观看HDR影像的视觉舒适度问题。此前有研究指出,如果在黑暗环境中观看高亮度HDR影像容易引发观众视觉不适的症状,在观看环境中提供适当的环境光照明,可以缓解这种视觉不适症状。本文通过视觉舒适度量化评价方法,研究HDR影像画面平均亮度、画面亮度分布和镜头间亮度跳跃等因素对视觉舒适度的影响,以及HDR影像亮度与环境光照度的关系,给出适宜的环境光照度范围,为HDR影像制作和播放环境配置提供参考。

基于4K超高清HDR技术的电视制作研究

在各类互联网平台高度发达的今天,人们的娱乐方式更加多样化,传统电视媒体面临观众流失的风险。不断提升电视节目的制作质量,增强电视媒体的市场竞争力,是传统电视媒体在传媒市场上立足和发展的应有之义。利用4K超高清高动态范围(High Dynamic Range,HDR)技术制作电视节目,视频的垂直、水平分辨率都可大幅提升,还能改变视频的亮度、色域及伽马值,全面提高视频及图像质量,进而提高电视节目的制作质量。基于此,首先介绍4K超高清HDR技术,其次分析4K HDR视频制作流程、播出流程,最后阐述4K超高清HDR技术在电视制作中的应用及4K HDR视频制作、播放要注意的问题,以供参考。

极端曝光图像的多尺度分解细节感知融合算法

针对极端曝光(欠曝光和过曝光)图像动态范围低的问题,提出一种基于多尺度分解细节感知的图像融合算法。欠曝光图像经过细节增强后,与过曝光图像进行初步的粗融合;采用小波变换对细节增强后图像的亮度分量进行多尺度分解,并设计专门的高频和低频融合策略,实现亮度分量的精融合;将粗融合图像的色调、饱和度分量,与精融合图像的亮度分量重组,获得最终的融合结果。基于大量测试数据,实验结果表明所提方法在视觉效果方面表现优异,平均MEF-SSIM指标为0.9854,平均SSIM指标为0.6508,均优于现有主流算法。

高动态范围条纹结构光在机检测技术及应用进展

条纹结构光技术是近年来发展迅速的非接触式测量方法,为机械加工在机检测提供了新的解决方案。由于加工环境光线复杂且金属零件本身具有高反光的特性,造成结构光在机检测的精度降低。将高动态范围(High Dynamic Range,HDR)技术应用于结构光检测中,可抑制高反光的影响,实现金属零件在复杂场景的测量。本文首先介绍了结构光测量原理,总结出HDR结构光在机检测面临的难点;其次,对HDR结构光技术进行了全面综述,以机械加工在机检测为背景,对基于硬件设备的HDR技术和基于条纹算法的HDR技术分别进行了归纳分析;然后,根据在机检测的条件需求,对各类技术进行总结,并比较不同方法的优缺点和在机检测的适用性;最后,结合近年来先进制造技术和精密测量的研究热点,对潜在应用进行分析,提出技术展望。

基于空间光调制器的HDR图像获取方法

提出一种利用空间光调制器对物方光线进行调制并多次曝光的方法来获取HDR图像.讨论了该方法的原理,并根据原理搭建实验装置进行了实验.实验装置中以LCOS作为空间光调制器,以CMOS作为成像传感器,通过多次曝光获得系列图像,利用图像合成算法生成HDR图像.该方法具有实时性好、扩展性强、成本较低等优点.

基于自适应多曝光融合的高动态范围表面测量方法

为解决结构光测量高动态范围表面物体时出现局部过度曝光或曝光不足的问题,提出一种改进的多曝光融合方法,利用自适应曝光代替手动曝光,并对图像融合过程进行优化。首先,将初始曝光时间下拍摄的图像利用直方图进行分析,将被测物体表面反射率不同的区域分为若干组,分别计算出每个组别的最佳曝光时间;在此基础上,拍摄不同组别对应最佳曝光时间下投射白光和条纹的图像,并去除图像中超过设定阈值的高灰度值区域,再将投射白光处理后的图像制作成掩模图,与相同曝光时间下投射条纹处理后的图像相乘,进而对多组相乘后的图像进行亮度压缩与融合;最后,通过CLAHE算法提高融合后所生成条纹图的对比度与清晰度,并对条纹解相后进行点云重建和尺寸测量。实验结果表明:文中方法中自适应曝光相较于手动曝光具有高效性和准确性,U型卡、连接块、圆盘三个高动态范围表面物体的点云重建率分别高达99.98%、99.74%、99.76%,测量出的标准块阶梯高度差绝对误差为0.062 mm,相对误差仅为0.69%,该方法有效解决了高动态范围表面物体测量时点云缺失的问题,提高了三维轮廓的测量精度。