Natural scene text detection based on YOLO V2 network model

Object detection could be a technology associated with computer vision and image process that deals with crime squad work instances of semantics objects of a specific category in digital pictures and videos. Applications of object detection are in various fields of computer vision, together with image retrieval. In the existing work, SVM model has been applied to sight the article in CCTV videos, unlike many other classification or detection problems there is a strong real-time requirement for detecting. Hence, a trade-off between high accuracy and speed is inescapable. The main parameters that influence the performance are the length of the feature vector along with the algorithmic programme, and also, it is the best in two classifications. The proposed system introduces You Only Look Once (YOLO V2). YOLO V2 is highly efficient and quicker than the SVM. The simulated results set the threshold level to any confidence when compared to SVM.KeywordsObject detectionSVMYOLO V2Image processingDeep learning

Efficient and adaptive computer vision systems have been proposed to make computer vision tasks, such as image classification and object detection, optimized for embedded or mobile devices. These solutions, quite recent in their origin, focus on optimizing the model (a deep neural network) or the system by designing an adaptive system with approximation knobs. Despite several recent efforts, we show that existing solutions suffer from two major drawbacks. First , while mobile devices or systems-on-chips usually come with limited resources including battery power, most systems do not consider the energy consumption of the models during inference. Second , they do not consider the interplay between the three metrics of interest in their configurations, namely, latency, accuracy, and energy. In this work, we propose an efficient and adaptive video object detection system— Virtuoso , which is jointly optimized for accuracy, energy efficiency, and latency. Underlying Virtuoso is a multi-branch execution kernel that is capable of running at different operating points in the accuracy-energy-latency axes, and a lightweight runtime scheduler to select the best fit execution branch to satisfy the user requirement. We position this work as a first step in understanding the suitability of various object detection kernels on embedded boards in the accuracy-latency-energy axes, opening the door for further development in solutions customized to embedded systems and for benchmarking such solutions. Virtuoso is able to achieve up to 286 FPS on the NVIDIA Jetson AGX Xavier board, which is up to 45× faster than the baseline EfficientDet D3 and 15× faster than the baseline EfficientDet D0. In addition, we also observe up to 97.2% energy reduction using Virtuoso compared to the baseline YOLO (v3)—a widely used object detector designed for mobiles. To fairly compare with Virtuoso , we benchmark 15 state-of-the-art or widely used protocols, including Faster R-CNN (FRCNN) [NeurIPS’15], YOLO v3 [CVPR’16], SSD [ECCV’16], EfficientDet [CVPR’20], SELSA [ICCV’19], MEGA [CVPR’20], REPP [IROS’20], FastAdapt [EMDL’21], and our in-house adaptive variants of FRCNN+, YOLO+, SSD+, and EfficientDet+ (our variants have enhanced efficiency for mobiles). With this comprehensive benchmark, Virtuoso has shown superiority to all the above protocols, leading the accuracy frontier at every efficiency level on NVIDIA Jetson mobile GPUs. Specifically, Virtuoso has achieved an accuracy of 63.9%, which is more than 10% higher than some of the popular object detection models, FRCNN at 51.1% and YOLO at 49.5%.

Text in images provides rich and accurate high-level semantic information, which is of great significance for many potential applications such as scene understanding, image and video retrieval. As a necessary means of understanding scene text, text detection has been widely concerned in the field of computer vision in recent years. However, due to the diversity of text scale and the uncertainty of image quality, text detection in natural scenes is still a challenging problem. Inspired by the successful application of deep neural network in target detection task, this paper proposes a text detection method based on multi-level feature extraction combined with deep learning technology. The multi-level feature extraction designed by this method can capture the details of the text, which is conducive to enhance the robustness of the algorithm to small-scale text and low-resolution images. In addition, the method is end-to-end, and can output text detection results directly. This paper evaluates the method on several standard datasets, and the experimental results prove the effectiveness of the proposed method.

As the use of digital media increases, effective retrieval and management techniques become more important. In order to extract useful information from this huge amount of data, many content based image retrieval (CBIR) systems have been developed. A typical CBIR system captures image features such as color, texture, or shape of objects in the query image and tries to retrieve images from the database with similar features. The main advantage of CBIR systems with relevance feedback is that these systems take into account the gap between the high-level concepts and low-level features and subjectivity of human perception of visual content. In this paper, a CBIR system based on color histogram has been presented and a new approach for image storage and retrieval called association-based image retrieval (ABIR) has been studied. The results of both systems have been shown.

The growing capacity of computers, the abundance of digital cameras and the increased connectivity of the world all point to large digital multimedia archives. They include images and videos from the World Wide Web, museum objects, flowers, trademarks, and views from everyday life. The faster they grow, the more prominently needed is the efficient access to the content of the images and videos. In this paper we have given important step of feature extraction, will be discussed in detail such as color, shape and texture information, particularly paying attention to discriminatory power and invariance. Then, we focus on the concepts of indexing and genre classification as intermediate step to sort the data. We pay attention to (interactive) ways to perform browsing and retrieval by means of information visualization and relevance feedback. Methods are being discussed to localize the retrieved objects in images. We adopt a hybrid approach for such text extraction by exploiting a number of characteristics of text blocks in color images and video frames. Our system detects both caption text as well as scene text of different font, size, color and intensity. Such texts are used for retrieval of video clips based on any given keyword. Content-Based Image And Video Retrieval addresses the basic concepts and techniques for designing content-based image and video retrieval systems.

Medical imaging is becoming a vital component of war on cancer. Tremendous amounts of medical image data are captured and recorded in a digital format during cancer care and cancer research. Facing such an unprecedented volume of image data with heterogeneous image modalities, it is necessary to develop effective and efficient content-based medical image retrieval systems for cancer clinical practice and research. While substantial progress has been made in different areas of content-based image retrieval (CBIR) research, direct applications of existing CBIR techniques to the medical images produced unsatisfactory results, because of the unique characteristics of medical images. In this paper, we develop a new multimodal medical image retrieval approach based on the recent advances in the statistical graphic model and deep learning. Specifically, we first investigate a new extended probabilistic Latent Semantic Analysis model to integrate the visual and textual information from medical images to bridge the semantic gap. We then develop a new deep Boltzmann machine-based multimodal learning model to learn the joint density model from multimodal information in order to derive the missing modality. Experimental results with large volume of real-world medical images have shown that our new approach is a promising solution for the next-generation medical imaging indexing and retrieval system.

Scene Text Deblurring in Non-stationary Video Sequences

This paper details about training a convolutional neural network for object detection and classification of a custom generated dataset from google earth satellite images. The objects of interest in satellite images are strategic targets such as nuclear power plants and oil refineries. The deep learning network is YOLO version 3 which has shown significant improvement in detecting smaller objects. YOLO v3 is three times faster than SSD and its AP metric for COCO dataset is on par with SSD. Hence YOLO v3 is a faster detector in comparison to SSD. But AP 0.75 is low when compared with RetinaNet due to higher localization error. On the other hand, the simpler network of YOLO, Tiny YOLO v3 takes lesser detection time. The objective of experimentation is to evaluate the performance of YOLO v3 and Tiny YOLO v3 for objects from satellite imagery. Google earth satellite images consumes less time and provides cost effective solution rather than acquiring the overhead images through unmanned aerial vehicles (UAVs) and drones. The objects size varies from 120 pixels to 1250 pixels. The experimental results demonstrate its detection capability, metrics results. GeForce RTX GPU was used for training the network.

The advancement in the broader field of Computer Vision is consequential, through past few decades. Therefore, a considerable improvement in object detection and tagging using convolutional neural networks has given way to accurate yet complex methods, which can identify objects in real-time. However, the growth in the area of implementing the algorithms on low powered portable devices has been relatively slow. This paper aims to converge the fields of computer vision and robotics, focusing on implementation of image description applications on an embedded system platform. We aim to integrate Neural Network powered object recognition system ‘YOLO v2’ with a robotic platform to explore the potential applications in the advancing domain of service and personal robotics.

Early detection and identification of plant diseases are still exceedingly difficult to do in the agriculture industry, despite their importance. We can efficiently accomplish this goal with the aid of Deep learning, an application of AI. In this study, the dataset contains 2310 images for the analysis and disease detection. For the diagnosis purpose, the ResNet 152V2, YOLO V3, and YOLO V5 were implemented on dataset. The whole image samples are further categorized into 3 major categories sets such as test, train, and validation. The images were preprocessed in the context of their identical size [224,224] for better feature selection process. After implementing the ResNet 152V2, YOLO V3, and YOLO V5 models, it can be seen that ResNet152V2 produces 26.13% Train-loss, 3S.22% Vald-loss, 93.54%, and S9.62% train-Acc and Vald-Acc respectively. Through YOLOV3, 21.20%, 96.41% was Train-loss and Vald-loss and 4S.S6%, 93.40% training and Vald-Acc was achieved. At the end of the YOLO V5 achieved higher accuracy (training and validation) i.e. 96.46%, 95.28% respectively and lesser the losses (Training and Validation) i.e. 19.57%, 19.00% respectively.

In this paper, we present a novel approach for medical content based image retrieval (searching for images which are pathologically similar to a given example) and demonstrate its performance on a dataset containing maxillofacial lesions. Recently, distributed databases at hospitals and CT scanning centers are related through Picture Archiving and Communication Systems (PACS). Hence, a content based image retrieval system is helpful for radiologists in medical diagnosis. In our proposed framework, a feature vector is extracted for each image and SIFT sparse codes are employed in this step. After feature extraction based on sparse coding and maximum pooling, we have utilized different similarity measures such as Euclidean norm, Manhattan distance and SVM classifier to choose most relevant images to the query image. We have evaluated our proposed framework on a dataset containing 2023 images belonging to 5 different categories: cleft, impaction, fracture, maxillary sinus cyst and healthy. Classification rate of 97.3% and precision versus recall curve indicate the effectiveness of sparse coding in content-based medical image retrieval.

Several Deep Learning (DL) and medical image Machine Learning (ML) methods have been investigated for efficient data representations of medical images, such as image classification, Content-Based Image Retrieval (CBIR), and image segmentation. CBIR helps medical professionals make decisions by retrieving similar cases and images from electronic medical image databases. CBIR needs expressive data representations for similar image identification and knowledge discovery in massive medical image databases explored by distinct algorithmic methods. In this study, an Improved Whale Optimization Algorithm with Deep Learning-Driven Retinal Fundus Image Grading and Retrieval (IWOADL-RFIGR) approach was developed. The presented IWOADL-RFIGR method mainly focused on retrieving and classifying retinal fundus images. The proposed IWOADL-RFIGR method used the Bilateral Filtering (BF) method to preprocess the retinal images, a lightweight Convolutional Neural Network (CNN) based on scratch learning with Euclidean distance-based similarity measurement for image retrieval, and the Least Square Support Vector Machine (LS-SVM) model for image classification. Finally, the IWOA was used as a hyperparameter optimization technique to improve overall performance. The experimental validation of the IWOADL-RFIGR model on a benchmark dataset exhibited better performance than other models.

The non-maximum suppression (NMS) algorithm, which merges neighboring bounding boxes as the detection result, is widely utilized in object detection methods. However, the traditional NMS algorithm is not suitable for text detection in natural scenes, especially for long texts and dense texts. In this paper, we observe that the coordinates of candidate bounding boxes are presented in skew distribution. On this observation, an improved NMS algorithm called SD-NMS (Skew Distribution NMS) is designed. First, the mode and the median of the coordinate set are counted to filter out the redundant bounding boxes. Then the left bounding boxes are merged for the location of the text regions. The SD-NMS method improves the detection ability of the model without extra model training and can be easily embedded in text detection methods. The experimental results show that our method obtains F-measure more than other NMS methods in public data sets ICDAR2015 and MSRATD500.

In the current ages, the progress in computer knowledge and multimedia solicitations has commanded to the construction of massive digital images and huge image databanks, and it is growing speedily. Here are numerous dissimilar expanses in which image retrieval shows a critical part resembling Medical systems, Forensic Labs, Tourism Elevation, etc. Thus, repossession of comparable images is an experiment. To challenge this speedy development in digital causes, it is essential to advance content-based image retrieval (CBIR) schemes, which can function on great databanks. Intelligent or deep learning way of content-based penetrating is requisite to accomplish the examining appeal with correct pictorial substances in a sensible quantity of time. There are some actually smart methods planned by investigators for effectual and vigorous content-based image retrieval. With these approaches, the quantity of approaches has been altered for the effectual image recovery of images. In this article, the review of various methods that have been utilized beginning from Image retrieval using pictorial structures like Bayesian Learning Algorithm, Self-Organizing Maps, Decision Trees Relevance Feedback, Genetic Programming, Navigation Pattern Mining, Association-Based Image Retrieval, Artificial Neural Networks (ANN) methodology is proposed and utilized for the content-based image retrieval. And the latest techniques such as the deep learning, ensemble learning methods with more number of layers and present suits the finest technique for repossession of images from the excessive databanks. In this effort, the goal is to best part the exertions of investigators who accompanied some strong exertion and to deliver an impervious perception for intellectual content-based image retrieval methods.KeywordsImage retrievalImage data storeTexture basedColor basedShape basedFeature extractionArtificial neural networksAnd deep learning

In recent years, secure image archival and retrieval model in cloud computing (CC) received significant attention to ensure data confidentiality and secured data transmission among cloud server and clients cloud storage and users. Traditional secure image retrieval (IR) techniques are not applicable to adapt with a large-scale IR in cloud environment. In order to overcome these issues, this article introduces an effective secure IR scheme using Inception with ResNet v2 (SIRS-IR) and Multiple Share Creation (MSC) scheme. The proposed method uses Inception with ResNet v2 model based feature extraction process. At the same time, the MSC process is utilized for multiple share generation and then encryption of shares takes place using double chaotic logistic map (DLCM) technique. Besides, the encrypted shares and the respective feature vectors are saved in the cloud server with the corresponding image identification number. During the IR process for the applied QI, the SIRS-IR model extracts the feature vectors and performs similarity measurement to retrieve the related images from the database interms of encrypted shares. Finally, the share decryption process is carried out for the reconstruction of original images with no loss of quality. Extensive experimentations were performed to verify the retrieval performance and image quality of the reconstructed images using Corel10K dataset. The obtained results stated that the presented SIRS-IR model is found to be superior to other methods in a considerable way.