Huge Image Data Research Articles

Review of Convolutional Neural Network Models and Image Classification

With the arrival of the era of big data and the improvement of computing power, deep learning has swept the world. Traditional image classification methods are difficult to deal with the huge image data, and cannot meet the requirements of people on the accuracy and speed of image classification, the image classification method based on convolutional neural network breaks through the bottleneck of the traditional image classification method, and becomes the mainstream algorithm of image classification, how to effectively use convolutional neural network to classify images has become a hot spot of research in the field of computer vision at home and abroad. In this paper, we review the research background, significance and current research status of convolutional neural network model and image classification, study two image classification methods based on ResNet and ShuffleNet, and provide a comprehensive review of the construction methods and characteristics of the two deep convolutional neural network model structures, and finally compare and analyse the performance of the two classification models.

Compact multi-spectral-resolution Wynne-Offner imaging spectrometer with a long slit.

Current imaging spectrometers are developed towards a large field of view (FOV) as well as high resolution to obtain more spatial and spectral information. However, imaging spectrometers with a large FOV and high resolution produce a huge image data cube, which increases the difficulty of spectral data acquisition and processing. In practical applications, it is more reasonable and helpful to identify different targets within a large FOV with different spectral resolutions. In this paper, a compact multi-spectral-resolution Wynne-Offner imaging spectrometer with a long slit is proposed by introducing a special diffraction grating with multi-groove densities at different areas. With the increasing of the groove density and the slit length, the astigmatism of the Wynne-Offner imaging spectrometer increases sharply. Therefore, the relationships between the astigmatism and both the groove density and slit length are studied. Moreover, a holographic grating is introduced. The holographic aberrations produced are utilized to balance the residual astigmatism of the imaging spectrometer. The design results show that the system is only 60m m×115m m×103m m in volume but achieves both a long slit of 20mm in length and a waveband from 400nm to 760nm with three kinds of spectral resolutions of 2nm, 1nm, and 0.5nm. The designed compact multi-spectral-resolution Wynne-Offner imaging spectrometer can be widely applied in the fields of crop classification and pest detection, which require both a large FOV and multiple spectral resolutions.

Review of Medical Image Segmentation Methods

Medical image is one of the key factors in the process of medical diagnosis and treatment. By analyzing the medical image obtained, doctors make judgments on the patient's condition and plan the next treatment process. Medical image segmentation is the process of segmenting areas of interest in medical images according to specific needs, which is a key step in medical image processing and analysis. With the great improvement of computer processing power, how to quickly and effectively segment huge image data and mining valuable information is the research hotspot of segmentation algorithm at present. This paper introduces several traditional medical image segmentation methods such as threshold method and region method, as well as convolutional neural network and Transformer segmentation method based on deep learning technology. Finally, this paper makes a summary and prospects for future development.

Pre-trained convolution neural networks models for content-based medical image retrieval

Content-based image retrieval (CBIR) is a recent method used to retrieve different types of images from repositories. The traditional content-based medical image retrieval (CBMIR) methods commonly used low-level image representation features extracted from color, texture, and shape image descriptors. Since most of these CBMIR systems depend mainly on the extracted features, the methods used in the feature extraction phase are more important. Features extraction methods, which generate inaccurate features, lead to very poor performance retrieval because of semantic gap widening. Hence, there is high demand for independent domain knowledge features extraction methods, which have automatic learning capabilities from input images. Pre-trained deep convolution neural networks (CNNs), the recent generation of deep learning neural networks, could be used to extract expressive and accurate features. The main advantage of these pre-trained CNNs models is the pre-training process for huge image data of thousands of different classes, and their knowledge after the training process could easily be transferred. There are many successful models of pre-trained CNNs models used in the area of medical image retrieval, image classification, and object recognition. This study utilizes two of the most known pre-trained CNNs models; ResNet18 and SqueezeNet for the offline feature extraction stage. Additionally, the highly accurate features extracted from medical images are used for the CBMIR method of medical image retrieval. This study uses two popular medical image datasets; Kvasir and PH2 to show that the proposed methods have good retrieval results. The retrieval performance evaluation measures of our proposed method have average precision of 97.75% and 83.33% for Kvasir and PH2 medical images respectively, and outperform some of the state-of-the-art methods in this field of study because these pre-trained CNNs have well trained layers among a huge number of image types. Finally, intensive statistical analysis shows that the proposed ResNet18-based retrieval method has the best performance for enhancing both recall and precision measures for both medical images.

A Survey of Image Classification Algorithms based on Convolution Neural Network

With the deep learning (DL) sweeping the world. Traditional image classification methods are difficult to process huge image data and cannot meet people's requirements for image classification accuracy and speed. The image classification method based on convolutional neural network (CNN) breaks through the bottle neck of traditional image classification methods and becomes the mainstream algorithm of image classification at present, how to effectively use convolutional neural network to classify images has become a hot research topic in the field of computer vision at home and abroad. Convolutional neural network (CNN) has performed well in image classification and segmentation, target detection and other applications, and its powerful feature learning and feature expression capabilities are increasingly respected by researchers. However, CNN still has a few problems, such as incomplete feature extraction and overfitting of sample training. In view of these problems, after in-depth research on the application of convolutional neural network in image processing, this paper gives the mainstream structure model, advantages and disadvantages, time/space complexity, problems that may be encountered in the model training process and corresponding solutions used in image classification based on convolutional neural network. Through the overview of the research status of CNN model in image classification, it provides suggestions for the further development and research direction of CNN.

Selective Medical Image Encryption Using Polynomial-Based Secret Image Sharing and Chaotic Map

The progressive development in telecommunication and networking technologies has led to the increased popularity of telemedicine usage, which involves the storage and transfer of medical images and related information. Recently, trust and privacy in the telemedicine system have attracted many researchers to investigate these topics. In medical image applications, selective image encryption plays an important role as it reduces computational cost and time. Therefore, a safe and efficient selective image encryption algorithm is designed in this work. In particular, the predetermined region of the original image data is encrypted to reduce the encryption/decryption time and the computational complexity of processing the huge image data. The image processing techniques are used to divide the image into a region of interest (ROI) and a region of non-interest (RONI), and then the more important component of the ROI is encrypted using a polynomial-based secret image sharing (SIS) and a chaotic map system. These techniques produce a test image cipher that has good confusion and diffusion properties. The experimental result shows that the Polynomial-based SIS and the chaotic image encryption are effectively performed for diffusion and confusion, which are crucial for concealment. According to the security research findings, sensitive encryption and decryption systems are extremely reliant on any improvement in the key. The encryption solution is sufficiently broad to withstand brute force attacks. Thus, protection may become an issue during the transmission of medical images via a network.

An Efficient Road Surveillance Approach to Detect, Recognize & Tracking Vehicles Using Deep Learning Methods

In the current scenario on the increasing number of motor vehicles day by day, so traffic regulation faces many challenges on intelligent road surveillance and governance, this is one of the important research areas in the artificial intelligence or deep learning. Among various technologies, computer vision and machine learning algorithms have the most efficient, as a huge vehicles video or image data on road is available for study. In this paper, we proposed computer vision-based an efficient approach to vehicle detection, recognition and Tracking. We merge with one-stage (YOLOv4) and two-stage (R-FCN) detectors methods to improve vehicle detection accuracy and speed results. Two-stage object detection methods provide high localization and object recognition precision, even as one-stage detectors achieve high inference and test speed. Deep-SORT tracker method applied for detects bounding boxes to estimate trajectories. We analyze the performance of the Mask RCNN benchmark, YOLOv3 and Proposed YOLOv4 + R-FCN on the UA-DETRAC dataset and study with certain parameters like Mean Average Precisions (mAP), Precision recall.

An Image Detection Method Based on Parameter Optimization of Support Vector Machine

With the progress and development of multimedia image processing technology, and the rapid growth of image data, how to efficiently extract the interesting and valuable information from the huge image data, and effectively filter out the redundant data, these have become an urgent problem in the field of image processing and computer vision. In recent years, as one of the important branches of computer vision, image detection can assist and improve a series of visual processing tasks. It has been widely used in many fields, such as scene classification, visual tracking, object redirection, semantic segmentation and so on. Intelligent algorithms have strong non-linear mapping capability, data processing capacity and generalization ability. Support vector machine (SVM) by using the structural risk minimization principle constructs the optimal classification hyper-plane in the attribute space to make the classifier get the global optimum and has the expected risk meet a certain upper bound at a certain probability in the entire sample space. This paper combines SVM and artificial fish swarm algorithm (AFSA) for parameter optimization, builds AFSA-SVM classification model to achieve the intelligent identification of image features, and provides reliable technological means to accelerate sensing technology. The experiment result proves that AFSA-SVM has better classification accuracy and indicates that the algorithm of this paper can effectively realize the intelligent identification of image features.

A Survey Paper on Image Mining Techniques and Classification Brain Tumor

Image mining is a method of searching and discovering the valuable information and knowledge from a set of huge image data. Image mining essentially depends on the data mining, digital image processing, machine learning, image retrieval, and artificial intelligence. Image mining is a process which is conducted to extract the hidden information such as image data and the additional pattern that could not be observed from the image. The main problems could face the mining of the collected images can be summarized in two main points: first is the image must be suitable for the mining process and second is the image’s chosen objects and features in order to be treated to extract the most effective route to save the time, and to save the effort. This paper is a survey presents the steps of the image mining process and represented an intensive view on using the image mining to the classify the brain tumors. In addition, it’s proposed a general scheme to accomplish the processes and to analysis the latest techniques which have been used to classify the brain tumors with comparison to the training groups and the amount of accuracy that obtained from the analysis. In addition, the paper compares the relevant and most recent published literature. The high published accuracy claim to be 98% which was obtained using the Deep convolutional neural network (DCNN).

Selective Image Encryption Based on DCT, Hybrid Shift Coding and Randomly Generated Secret Key

Most of today’s techniques encrypt all of the image data, which consumes a tremendous amount of time and computational payload. This work introduces a selective image encryption technique that encrypts predetermined bulks of the original image data in order to reduce the encryption/decryption time and thecomputational complexity of processing the huge image data. This technique is applying a compression algorithm based on Discrete Cosine Transform (DCT). Two approaches are implemented based on color space conversion as a preprocessing for the compression phases YCbCr and RGB, where the resultant compressed sequence is selectively encrypted using randomly generated combined secret key.The results showed a significant reduction in image quality degradation when applying the system based on YCbCr over RGB, where the compression ratio was raised in some of the tested images to 50% for the same Peak Signal to Noise Ratio (PSNR). The usage of 1-D DCT reduced the transform time by 47:1 times comparedto the same transform using 2-D DCT. The values of the adaptive scalar quantization parameters were reduced to the half for the luminance (Y band) to preserve the visual quality, while the chrominance (Cb and Cr bands) were quantized by the predetermined quantization parameters. In the hybrid encoder horizontal zigzag,block scanning was applied to scan the image. The Detailed Coefficient (DC) coefficients are highly correlated in this arrangement- where DC are losslessly compressed by Differential Pulse Coding Modulation (DPCM) and theAccumulative Coefficients (AC) are compressed using Run Length Encoding (RLE). As a consequence, for the compression algorithm, the compression gain obtained was up to 95%. Three arrays are resulted from each band (DC coefficients, AC values, and AC runs), where the cipher is applied to some or all of those bulksselectively. This reduces the encryption decryption time significantly, where encrypting the DC coefficients provided the second best randomness and the least encryption/decryption time recorded (3 10-3 sec.) for the entire image. Although the compression algorithm consumes time but it is more efficient than the savedencryption time.

A Novel Framework design for Semantic based Image retrieval as a Cyber Forensic Tool

Cyber forensics includes the areas of computer forensics, network forensics and internet forensics. Digital images were commonly used in cyber forensics to collect criminal images, fingerprints, images of crime events and so on. Since the current cyber forensic tools are not very much furnished with the course of action of huge image data, it becomes a big issue to obtain image evidence to prosecute the criminal, most of the evidence is available in the form of raw semantics. Cyber forensic investigators often face the challenge of manually examining an enormous quantity of digital image information to identify direct evidence with the assistance of this semantics. Semantic based image retrieval system (SBIR) is therefore the recent and best option to solve this drawback. The main purpose of this research is to design for cyber forensic tools a novel framework of semantic-based image retrieval system. We therefore present a deep learning framework based on the Convolution neural network(GoogLeNet) for recognizing distinct facial expressions from the Yale facial image database for cyber associated forensic tools, the presented framework is very effective for classification and detection based on semantics or verbal descriptions. The network has accomplished a decent accuracy of 86.25 % after training.

Parallel Processing of Image Segmentation Data Using Hadoop

The use of sequential programming is slowly getting replaced by distributed and parallel computing which is widely being used in computing industries to handle tasks with big data and various high-end computing applications comprising of huge image and video data banks. Moreover, image processing using parallel computation is also gaining momentum in today's technological era. Nowadays researchers are coming up with various methodologies to tackle high scale image processing applications by implementing parallel computing methodologies to carry out the specified image processing application task and simultaneously checking its performance against sequential programming. At the same time there are constraints on what can be done to maximize the task performance using high end multi-core CPU's with advanced buses and interconnects that offer high bandwidth with low system latency. It is to be noted that there is no availability of standardized image processing task which can be used to evaluate a single node system. In this paper, we propose an efficient parallel processing algorithm to perform the task of image segmentation with the foremost aim to analyze the threshold of data size at which the proposed method outperforms sequential programming method in terms of task execution time by analyzing the distribution of average CPU cores usage and its threads over the execution time. The proposed methodology could be useful for researchers, as it can perform multiple image segmentation in parallel, which can save a lot of time of the user. For the purpose of comparison, we also implemented the same image segmentation task using sequential method of programming in an integrated development environment platform.

Quantitative Co-Localization and Pattern Analysis of Endo-Lysosomal Cargo in Subcellular Image Cytometry and Validation on Synthetic Image Sets.

Late endosomes and lysosomes (LE/LYSs) play a central role in trafficking of endocytic cargo, secretion of exosomes, and hydrolysis of ingested proteins and lipids. Failure in such processes can lead to lysosomal storage disorders in which a particular metabolite accumulates within LE/LYSs. Analysis of endocytic trafficking relies heavily on quantitative fluorescence microscopy, but evaluation of the huge image data sets is challenging and demands computer-assisted statistical tools. Here, we describe how to use SpatTrack ( www.sdu.dk/bmb/spattrack ), an imaging toolbox, which we developed for quantification of the distribution and dynamics of endo-lysosomal cargo from fluorescence images of living cells. First, we explain how to analyze experimental images of endocytic processes in Niemann Pick C2 disease fibroblasts using SpatTrack. We demonstrate how to quantify the location of the sterol-binding protein NPC2 in LE/LYSs relative to cholesterol -rich lysosomal storage organelles (LSOs) stained with filipin. Second, we show how to simulate realistic vesicle patterns in the cell geometry using Markov Chain Monte Carlo and suitable inter-vesicle and cell-vesicle interaction potentials. Finally, we use such synthetic vesicle patterns as "ground truth" for validation of two-channel analysis tools in SpatTrack, revealing their high reliability. An improved version of SpatTrack for microscopy-based quantification of cargo transport through the endo-lysosomal system accompanies this protocol.

Development of a 97.5-inch Super Hi-Vision 8 K × 4 K liquid crystal display panel

A prototype 97.5-inch 8 K×4 K a-Si liquid crystal display (LCD) panel for Super Hi-Vision with ultra-high definition was successfully developed using the BOE exclusive advanced super-dimensional switching technology. This LCD has a panel with 7680(×3) by 4320 pixels, 16 times those of the conventional Full High Definition (FHD), and the pixel size of the panel is 0.2811 mm (90.4 pixels per inch). The liquid crystal charging time was extended by quad area and double source line structure (also called “2G2D driving”). Additionally, a multi-functional image-processing video board was developed to compatibly realize the real-time display of huge image data from FHD to 8 K. Furthermore, the reason for the arcing that occurred in half-tone source–drain mask etching was discussed and was ultimately solved by improving the exposure process. The demonstrated panel features a 1200:1 contrast ratio, 500 nit brightness, and 72% color gamut.

The Image Compression Technology Based on Wavelet Transform

Image compression is a data compression technology used in the digital image, its purpose is to reduce redundant information of the image data, and provide a more efficient format to store and transmit data. Due to the huge image data and the existing relatively low transport conditions, the image compression has become an inevitable. The key technology of image compression is how to transform image data, how to quantify image data, and how to entropy code the quantized data. Using two-dimensional Mallat image wavelet compression algorithm is a new method of image compression, and it is the core technology of the wavelet image compression.

Behavior Identification Based on Geotagged Photo Data Set

The popularity of mobile devices has produced a set of image data with geographic information, time information, and text description information, which is called geotagged photo data set. The division of this kind of data by its behavior and the location not only can identify the user's important location and daily behavior, but also helps users to sort the huge image data. This paper proposes a method to build an index based on multiple classification result, which can divide the data set multiple times and distribute labels to the data to build index according to the estimated probability of classification results in order to accomplish the identification of users' important location and daily behaviors. This paper collects 1400 discrete sets of data as experimental data to verify the method proposed in this paper. The result of the experiment shows that the index and actual tagging results have a high inosculation.

GPU-Accelerated Discrete Wavelet Transform for Images

Discrete Wavelet Transform (DWT) has been brought into wide use in image processing, but it cant fit the demand of the hugeimage data because the time of computing is vast. The GPU is an attractive platform for a broad fieldof applications,which remains asignificanthigharithmetic processingcapability. Therefore itcan beusedasa powerful accelerator without extra cost.CUDA(computeunifieddevicearchitecture) providesahardwareandsoftwareenvironment touse the GPU to accelerate the DWT for images. In this paper, we use the NVIDIA GeForce GT 650M that complies with the CUDA to improvethe execution time of theDiscrete Wavelet Transformfor images. TheresultofexperimentsindicatesthattheCUDAtechnology hastheadvantagesof parallel processingandtheefficiencyofimagetransform isimprovedgreatly. Whats more, it performs better on the larger size image (the max speedup is 15.9).

LCD 시스템을 위한 Current-Mode Multi-Valued Logic 인터페이스 회로

In this paper, we propose interface circuits for reducing power consumption and EMI when sequences of data from LCD controller to LCD driver IC by transmitting two bit data during one clock period. The proposed circuits are operated in current mode, which is different from conventional voltage-mode signaling techniques, and also employ threshold technique of Modified-LVDS(Low Voltage Differential Signaling) method. We have simulated the proposed circuits using H-SPICE tool for performance analysis of the proposed method. The simulation results show that the proposed circuits provide a faster transmission speed and stronger noise immunity than the conventional LVDS circuits. It might be suitable for the real-time transmission of huge image data in LCD system.

고성능 HEVC 복호기를 위한 효율적인 32×32 역변환기 설계

본 논문에서는 고성능 HEVC 복호기를 위한 효율적인 <TEX>$32{\times}32$</TEX> 역변환기 하드웨어 구조를 제안한다. HEVC는 4k, 8k 이미지와 같이 기존의 이미지코덱에 비해 훨씬 더 큰 크기의 이미지를 처리할 수 있는 새로운 영상 압축 표준이다. 큰 이미지의 데이터를 효과적으로 처리하기 위해 다양한 새 블록 구조를 채택하였으며, 이 블록들은 <TEX>$4{\times}4$</TEX>, <TEX>$8{\times}8$</TEX>, <TEX>$16{\times}16$</TEX>, <TEX>$32{\times}32$</TEX>으로 구성되었다. 이 논문에서는 <TEX>$32{\times}32$</TEX> 역변환기의 효과적인 구조를 제안하며, 역변환기의 구조는 <TEX>$32{\times}32$</TEX> 행렬을 <TEX>$16{\times}16$</TEX> 행렬로 재구성하고 쉬프트와 덧셈기로 구성된 곱셈기를 사용하여 연산을 단순화 하였으며 멀티 사이클 패스를 구현하여 낮은 주파수에서도 동작이 가능하도록 설계하였다. 또한 HEVC 코덱의 다양한 크기의 변환이나 순방향 변환 블록에 쉽게 적용할 수 있다. In this paper, an efficient hardware architecture is proposed for <TEX>$32{\times}32$</TEX> inverse transform HEVC decoder. HEVC is a new image compression standard to deal with much larger image sizes compared with conventional image codecs, such as 4k, 8k images. To process huge image data effectively, it adopts various new block structures. Theses blocks consists of <TEX>$4{\times}4$</TEX>, <TEX>$8{\times}8$</TEX>, <TEX>$16{\times}16$</TEX>, and <TEX>$32{\times}32$</TEX> block. This paper suggests an effective structures to process <TEX>$32{\times}32$</TEX> inverse transform. This structure of inverse transform adopts the decomposed <TEX>$16{\times}16$</TEX> matrixes of <TEX>$32{\times}32$</TEX> matrix, and simplified the operations by implementing multiplying with shifters and adders. Additionally the operations frequency is downed by using multicycle paths. Also this structure can be easily adopted to a multi-size transform or a forward transform block in HEVC codec.

Semantic Learning and Web Image Mining with Image Recognition and Classification

Image mining is more than just an extension of data mining to image domain. Web Image mining is a technique commonly used to extract knowledge directly from images on WWW. Since main targets of conventional Web mining are numerical and textual data, Web mining for image data is on demand. There are huge image data as well as text data on the Web. However, mining image data from the Web is paid less attention than mining text data, since treating semantics of images are much more difficult. This paper proposes a novel image recognition and image classification technique using a large number of images automatically gathered from the Web as learning images. For classification the system uses imagefeature- based search exploited in content-based image retrieval(CBIR), which do not restrict target images unlike conventional image recognition methods and support vector machine(SVM), which is one of the most efficient &amp; widely used statistical method for generic image classification that fit to the learning tasks. By the experiments it is observed that the proposed system outperforms some existing search systems.