Conventional Image Processing Algorithms Research Articles

Deep learning and hybrid approach for particle detection in defocusing particle tracking velocimetry

The present work aims at the improvement of particle detection in defocusing particle tracking velocimetry (DPTV) by means of a novel hybrid approach. Two deep learning approaches, namely faster R-CNN and RetinaNet are compared to the performance of two benchmark conventional image processing algorithms for DPTV. For the development of a hybrid approach with improved performance, the different detection approaches are evaluated on synthetic and images from an actual DPTV experiment. First, the performance under the influence of noise, overlaps, seeding density and optical aberrations is discussed and consequently advantages of neural networks over conventional image processing algorithms for image processing in DPTV are derived. Furthermore, current limitations of the application of neural networks for DPTV are pointed out and their origin is elaborated. It shows that neural networks have a better detection capability but suffer from low positional accuracy when locating particles. Finally, a novel Hybrid Approach is proposed, which uses a neural network for particle detection and passes the prediction onto a conventional refinement algorithm for better position accuracy. A third step is implemented to additionally eliminate false predictions by the network based on a subsequent rejection criterion. The novel approach improves the powerful detection performance of neural networks while maintaining the high position accuracy of conventional algorithms, combining the advantages of both approaches.

Establishment of an antifouling performance index derived from the assessment of biofouling on typical marine sensor materials

Marine biofouling, known as the unwanted accumulation of living organisms on submerged surfaces, is one of the main factors affecting the operation, maintenance and data quality of water quality monitoring sensors. This can be a significant challenge for marine deployed infrastructure and sensors in water. When organisms attach to the mooring lines or other submerged surfaces of the sensor, they can interfere with the sensor's operation and accuracy. They can also add weight and drag to the mooring system, making it more difficult to maintain the desired position of the sensor. This increases the cost of ownership to the point where it becomes prohibitively expensive to maintain operational sensor networks and infrastructures.Furthermore, the analysis and quantification of biofouling is extremely complex as it is based on biochemical methods such as the analysis of pigments such as chlorophyll-a as a direct indicator of the biomass of photosynthetic organisms, dry weight, carbohydrate analysis and protein analysis among others.In this context, this study has developed a method to estimate biofouling quickly and accurately on different submerged materials used in the marine industry and specifically in sensor manufacturing like copper, titanium, fiberglass composite, different types of polyoxymethylene (POMC, POMH), polyethylene terephthalate glycol (PETG) and 316L-stainless steel. To do this, in situ images of fouling organisms were collected with a conventional camera and image processing algorithms and machine learning models trained were used to construct a biofouling growth model. The algorithms and models were implemented with Fiji-based Weka Segmentation software. A supervised clustering model was used to identify three types of fouling to quantify fouling on panels of different materials submerged in seawater over time. This method is easy, fast and cost-effective to classify biofouling in a more accessible and holistic way that could be useful for engineering applications.

Assessment of plant disease detection by deep learning

Plant disease and pest detection machines were originally used in agriculture and have, to some extent, replaced traditional visual identification. Plant diseases and pests are important determinants of plant productivity and quality. Plant diseases and pests can be identified using digital image processing. According to the difference in the structure of the network, this study presents research on the detection of plant diseases and pests based on three aspects of the classification network, detection network, and segmentation network in recent years, and summarizes the advantages and disadvantages of each method. A common data set is introduced and the results of existing studies are compared. This study discusses possible problems in the practical application of plant disease and pest detection based on deep learning. Conventional image processing algorithms or manual descriptive design and classifiers are often used for traditional computer vision-based plant disease and pest detection. This method usually uses various characteristics of plant diseases and pests to create an image layout and selects a useful light source and shooting angle to produce evenly lit images. The purpose of this work is to identify a group of pests and diseases of domestic and garden plants using a mobile application and display the final result on the screen of a mobile device. In this work, data from 38 different classes were used, including diseased and healthy leaf images of 13 plants from plantVillage. In the experiment, Inception v3 tends to consistently improve accuracy with an increasing number of epochs with no sign of overfitting and performance degradation. Keras with Theano backend used to teach architectures

DSRGAN: Detail Prior-Assisted Perceptual Single Image Super-Resolution via Generative Adversarial Networks

The generative adversarial network (GAN) is successfully applied to study the perceptual single image super-resolution (SISR). However, since the GAN is data-driven, it has a fundamental limitation on restoring real high frequency information for an unknown instance (or image) during test. On the other hand, the conventional model-based methods have a superiority to achieve instance adaptation as they operate by considering the statistics of each instance (or image) only. Motivated by this, we propose a novel model-based algorithm, which can extract the detail layer of an image efficiently. The detail layer represents the high frequency information of image and it is constituted of image edges and fine textures. It is seamlessly incorporated into the GAN and serves as a prior knowledge to assist the GAN in generating more realistic details. The proposed method, named DSRGAN, takes advantages from both the model-based conventional algorithm and the data-driven deep learning network. Experimental results demonstrate that the DSRGAN outperforms the state-of-the-art SISR methods on perceptual metrics, meanwhile achieving comparable results in terms of fidelity metrics. Following the DSRGAN, it is feasible to incorporate other conventional image processing algorithms into a deep learning network to form a model-based deep SISR.

EU‐Net: A novel semantic segmentation architecture for surface defect detection of mobile phone screens

Manual or conventional image processing algorithms are commonly used to detect surface problems on mobile phone screens. However, inefficiency and inflexibility are disadvantages. Although the semantic segmentation method has high adaptability and accuracy, it also has a low defect detection efficiency due to its excessive parameters. In order to increase defect detection efficiency, a novel efficient encoder–decoder architecture termed MB encoder–decoder architecture based on MBConv blocks, and that it reduces the number of parameters used in semantic segmentation methods i presented. In addition, by applying the MB encoder–decoder design to the U-Net, the efficient U-Net (EU-Net) is proposed. It confirms the MB encoder–decoder architecture's superiority. Then, EU-Net to mobile phone surface defect detection in real industrial scenarios. Experimental results on a dataset show the superiority of the proposed algorithm and it can meet the real-time requirement of industrial production.

Deep learning provides high accuracy in automated chondrocyte viability assessment in articular cartilage using nonlinear optical microscopy.

Chondrocyte viability is a crucial factor in evaluating cartilage health. Most cell viability assays rely on dyes and are not applicable for in vivo or longitudinal studies. We previously demonstrated that two-photon excited autofluorescence and second harmonic generation microscopy provided high-resolution images of cells and collagen structure; those images allowed us to distinguish live from dead chondrocytes by visual assessment or by the normalized autofluorescence ratio. However, both methods require human involvement and have low throughputs. Methods for automated cell-based image processing can improve throughput. Conventional image processing algorithms do not perform well on autofluorescence images acquired by nonlinear microscopes due to low image contrast. In this study, we compared conventional, machine learning, and deep learning methods in chondrocyte segmentation and classification. We demonstrated that deep learning significantly improved the outcome of the chondrocyte segmentation and classification. With appropriate training, the deep learning method can achieve 90% accuracy in chondrocyte viability measurement. The significance of this work is that automated imaging analysis is possible and should not become a major hurdle for the use of nonlinear optical imaging methods in biological or clinical studies.

Automotive Brake Part Inspection and Fault Localization using Deep Learning

Inspection of brake components is very essential to detect the damaged manufactured parts before it is assembled in any vehicle. Manual inspection of brakes is extremely difficult since most of defects are very minute and cannot be identified by human eyes. Therefore, automatic inspection of manufactured brakes is indispensible to prevent failure of brakes and accidents. Previously, various research articles perform inspection of brake through conventional image processing and traditional image processing algorithms. However, these techniques are capable of identifying a single fault only and are less robust to detecting numerous faults. Further, the existing techniques hardly localize the exact location of faults in the surface of brake. In order to over these drawbacks, in this research we utilize deep learning object detection algorithms namely Single Shot Detector and Faster RCNN to identify and localize the exact location of fault on the brake surface. Furthermore, the proposed system is capable to detect different types of faults in a single algorithm and is robust to brake’s material surface, environmental and lightening factors. The deep learning algorithms are trained using transfer learning on custom collected dataset. The proposed algorithms deliver an accuracy of 95.64% and mAP of 73.2% on cylindrical grey shade brakes.

A comprehensive review on soil classification using deep learning and computer vision techniques

Soil classification is one of the major affairs and emanating topics in a large number of countries. The population of the world is rising at a majorly rapid pace and along with the increase in population, the demand for food surges actively. Typical techniques employed by the farmers are not adequate enough to fulfill the increasing requirements and therefore they have to hinder the cultivating soil. For proper crop yield, farmers should be aware of the correct soil type for a particular crop, which affects the increased demand for food. There are various laboratory and field methods to classify soil, but these have limitations like time and labor-consuming. There is a requirement of computer-based soil classification techniques which will help farmers in the field and won’t take a lot of time. This paper talks about different computer-based soil classification practices divided into two streams. First is image processing and computer vision-based soil classification approaches which include the conventional image processing algorithms and methods to classify soil using different features like texture, color, and particle size. Second is deep learning and machine learning-based soil classification approaches, such as CNN, which yields state-of-the-art results. Deep learning applications mostly diminish the dependency on spatial-form designs and preprocessing techniques by facilitating the end-to-end process. This paper also presents some databases created by the researchers according to the objective of the study. Databases are created under different environmental and illumination conditions, using different appliances such as digital cameras, digital camcorder, and a smartphone camera. Also, evaluation metrics are briefly discussed to layout some graded measures for differentiation. This review serves as a brief guide to new researchers in the field of soil classification, it provides fundamental understanding and general knowledge of the modern state-of-the-art researches, in addition to skillful researchers considering some dynamic trends for future work.

Image processing algorithms for biospeckle analysis of almond seed

Laser biospeckle technique, as a non-invasive, non-contact and non-destructive method has gained importance in the assessment of biospeckle activity of various biological samples. This study presents the application of the laser biospeckle technique [1] to evaluate the activities of dry and soaked almond seed. We have used conventional image processing algorithms such as inertia moment (IM), absolute value of differences (AVD), and probability mass function of regular difference, temporal speckle contrast and fujii for this analysis. All analyses indicate an increase of biospeckle activity of the soaked almond (402.85 using IM) in comparisons to the dry stage (44.32 using IM) which can be attributed to the initiation of germination. Spectral activity maps obtained using fujii represents the change of surface of the almond seed after soaking water.

An attempt to detect anomalies in CT-data of car body parts using machine learning algorithms

Industries, which produce hundreds of terabyte of CT data per year, demand automated evaluation approaches. This work provides a first glance of an attempt to automatically detect and characterize possible defects and/or anomalies which formed during common joining processes. We investigated a standard riveting process with respect to the resulting final head height of steel selfpiercing half-hollow rivets. The methods include conventional image processing algorithms, like edge-detection, thresholding and principle component analysis (PCA) which were used to pre-process the CT data. In order to automatically evaluate the reconstructed volumes, which contained several of the aforementioned rivets, we compared the performance of different, publicly available, convolutional neural network (CNN) architectures. Furthermore, we investigated the impact of data augmentation and showed by means of a k-fold cross-validation that the training data causes no overfitting of the network. The obtained results suggest that an automated evaluation of the generated computed tomography scans, with regard to a rivet’s final head height, is feasible. However, in order to increase the network’s reliability and accuracy, the amount of training data needs to be further enlarged and diversified.

Hough Transform Implementation For Event-Based Systems: Concepts and Challenges.

Hough transform (HT) is one of the most well-known techniques in computer vision that has been the basis of many practical image processing algorithms. HT however is designed to work for frame-based systems such as conventional digital cameras. Recently, event-based systems such as Dynamic Vision Sensor (DVS) cameras, has become popular among researchers. Event-based cameras have a significantly high temporal resolution (1 μs), but each pixel can only detect change and not color. As such, the conventional image processing algorithms cannot be readily applied to event-based output streams. Therefore, it is necessary to adapt the conventional image processing algorithms for event-based cameras. This paper provides a systematic explanation, starting from extending conventional HT to 3D HT, adaptation to event-based systems, and the implementation of the 3D HT using Spiking Neural Networks (SNNs). Using SNN enables the proposed solution to be easily realized on hardware using FPGA, without requiring CPU or additional memory. In addition, we also discuss techniques for optimal SNN-based implementation using efficient number of neurons for the required accuracy and resolution along each dimension, without increasing the overall computational complexity. We hope that this will help to reduce the gap between event-based and frame-based systems.

수정된 Watershed 알고리즘 이용한 빠른 Mura 결함 검출

Many kinds of defects show up during the process of manufacturing display panels. However, mura defects are the most difficult to detect using the conventional image processing algorithms. Many factors cause mura defects to appear in display panels. When images are taken using cameras, mura defects normally show up as relatively dark or bright regions with no definite shape, no clear contours, and very low contrast against their surrounding background. When an imaged mura defect is relatively dark compared to its background, it can be considered a water catchment basin when the whole image is visualized in three dimensions (i.e., is topographically interpreted), and such catchment basins can be detected by watershed algorithms. In this paper, for the accurate segmentation of the mura region, the flooding step of the original watershed algorithm is carefully redesigned to detect the mura defect that exists both inside and at the boundary of an image. The depth of the catchment basins is recorded iteratively and then is used to segment the mura defects. The just noticeable difference (JND) technique is used to quantify the level of the mura defects. It is shown, by extensive experiments, that the proposed algorithm performs well, detecting very low-contrast mura defects, and quickly detects defects located anywhere in the image.

Novel motion estimation algorithm for image stabilizer

PurposeThe motion vector estimation algorithm is very widely used in many image process applications, such as the image stabilization and object tracking algorithms. The conventional searching algorithm, based on the block matching manipulation, is used to estimate the motion vectors in conventional image processing algorithms. During the block matching manipulation, the violent motion will result in greater amount of computation. However, too large amount of calculation will reduce the effectiveness of the motion vector estimation algorithm. This paper aims to present a novel searching method to estimate the motion vectors effectively.Design/methodology/approachThis paper presents a novel searching method to estimate the motion vectors for high-resolution image sequences. The searching strategy of this algorithm includes three steps: the larger area searching, the adaptive directional searching and the small area searching.FindingsThe achievement of this paper is to develop a motion vector searching strategy to improve the computation efficiency. Compared with the conventional motion vector searching algorithms, the novel motion vector searching algorithm can reduce the motion matching manipulation effectively by 50 per cent.Originality/valueThis paper presents a novel searching strategy to estimate the motion vectors effectively. From the experimental results, the novel motion vector searching algorithm can reduce the motion matching manipulation effectively, compared with the conventional motion vector searching algorithms.

An Object-Based Approach to Image/Video-Based Synthesis and Processing for 3-D and Multiview Televisions

This paper proposes an object-based approach to a class of dynamic image-based representations called ldquoplenoptic videos,rdquo where the plenoptic video sequences are segmented into image-based rendering (IBR) objects each with its image sequence, depth map, and other relevant information such as shape and alpha information. This allows desirable functionalities such as scalability of contents, error resilience, and interactivity with individual IBR objects to be supported. Moreover, the rendering quality in scenes with large depth variations can also be improved considerably. A portable capturing system consisting of two linear camera arrays was developed to verify the proposed approach. An important step in the object-based approach is to segment the objects in video streams into layers or IBR objects. To reduce the time for segmenting plenoptic videos under the semiautomatic technique, a new object tracking method based on the level-set method is proposed. Due to possible segmentation errors around object boundaries, natural matting with Bayesian approach is also incorporated into our system. Furthermore, extensions of conventional image processing algorithms to these IBR objects are studied and illustrated with examples. Experimental results are given to illustrate the efficiency of the tracking, matting, rendering, and processing algorithms under the proposed object-based framework.

Complete automatic target cuer/recognition system for tactical forward-looking infrared images

A complete forward-looking IR (FLIR) automatic target cuer/ recognizer (ATC/R) is presented. The data used for development and testing of this ATC/R are first generation FLIR images collected using a F-15E. The database contains thousands of images with various mission profiles and target arrangements. The specific target of interest is a mobile missile launcher, the primary target. The goal is to locate all vehicles (secondary targets) within a scene and identify the primary targets. The system developed and tested includes an image segmenter, region cluster algorithm, feature extractor, and classifier. Conventional image processing algorithms in conjunction with neural network techniques are used to form a complete ATC/R system. The conventional techniques include hit/miss filtering, difference of Gaussian filtering, and region clustering. A neural network (multilayer perceptron) is used for classification. These algorithms are developed, tested and then combined into a functional ATC/R system. Overall primary target detection rate (cuer) is 84% with a 69% primary target identification (recognizer) rate at ranges relevant to munitions release. Furthermore, the false alarm rate (a nontarget cued as a target) is only 2.3 per scene. The research is being completed with a 10 flight test profile using third generation FLIR images.

A neural network for image-based vehicle detection

Vehicle detection on roadways is useful for a variety of traffic engineering applications from intersection signal control to transportation planning. Traditional detection methods have relied on mechanical or electrical devices placed on top of, or embedded in, pavements. These systems are relatively expensive to install, tend to be unreliable over time, and are limited in their capabilities. Considerable research has been conducted in the area of machine vision for Wide Area Vehicle Detections Systems (WADS). These systems have typically employed conventional image processing and pattern matching algorithms, and many installations have been sensitive to varying lighting conditions, camera perspective, and shadows. In addition, these systems have often required large amounts of computing resources. This paper reports on the development of a new image based vehicle detection system that is based on a simple back propagation/feedforward neural network for tracking vehicles. Application of this concept in a field system is discussed and preliminary results are presented. These results suggest that the neural network vehicle tracking model can be used to reliably detect vehicles. In addition, the training capability of the neural network detection model permits the system to adapt to variations in lighting and camera placement. This should lead to simplified installation and maintenance of WADS.

Comparison of image processing algorithms and neural networks in machine vision inspection

Automated vision inspection has become a vital part of the computer integrated manufacturing systems. This paper compares the development and performance of two methodologies for a machine vision inspection system. The first method is developed through conventional image processing algorithms and the second method is based on the neural networks. A case study was conducted to benchmark these two methods. The results showed that the conventional image processing algorithms required less development time than the neural networks. A considerable amount of time was spend on training the neural networks. However, the neural networks performed better than the conventional image processing algorithms in terms of accuracy.