Frequency domain attention network for S/TEM image segmentation

In the optimization image segmentation process, the traditional algorithm is used for segmentation, because the calculation method is complex, image segmentation accuracy rate is low, the clarity is not high, when the amount of information in the image segmentation is very large, the time consumption is big. Aiming at the above problems, an improved image optimization segmentation method is proposed based on wavelet algorithm. The image feature is obtained based on wavelet transform, the image characteristics can be reacted in local information, the advantage is obvious, the adjacent scale wavelet coefficients multiplication method isused to remove the noise, and the wavelet algorithm and graph theory algorithm is fused to generate the minimum spanning tree, the accurate image segmentation is obtained with minimum spanning tree. The simulations show the method is applied in image segmentation, the precision is high, and it has strong applicability. Introduction In the research field of image, the technology of image segmentation is a kind of very important technology of image processing, it is widely applied in many aspects of image processing, the general idea of image segmentation technology is to put a large image to divideinto many small pieces, previous research shows the methods and types of image segmentation is great many. But there are no unified standard to measure, this has brought trouble tothe large-scale application. So how to establish a reasonable optimization of general image segmentation method is the key to solve the issue, more and more experts have put a lot ofrelated attention to the field. At this stage, the main methods of image segmentation include ant colony algorithm, image segmentation method based on pixel gray difference algorithm, and image segmentation method based on adaptive threshold segmentation algorithm. Among them, the most commonly used method is the image segmentation based on adaptive thresholdsegmentation algorithm. According to the traditional image segmentation methods, the image surface is core, this method is used in segmentation, it has strong operability, it is simple and it has obtained the broad application. But its disadvantage is that the surface is often leads to excessive segmentation, the segmentation performance is bad. The computational complexity is big[1]. Therefore, an improved image optimization segmentation method is proposed based on wavelet algorithm. The image feature is obtained based on wavelet transform, the image characteristics can be reacted in local information, the advantage is obvious, the adjacent scale wavelet coefficients multiplication method is used to remove the noise, and the wavelet algorithm and graph theory algorithm is fused to generate the minimum spanning tree, the accurate image segmentation is obtained with minimum spanning tree. The simulations show the method is applied in image segmentation, the precision is high, and it has strong applicability. Image segmentation method based on improved wavelet algorithm A Image feature extraction and denoising.The advantage of wavelet transform is that it can provide the time domain and frequency domain window with the image features. It has the ability of reaction of the local features of image information, so it can consider to extract local features of images by using the characteristics of wavelet transform, wavelet transform is used for image International Conference on Automation, Mechanical Control and Computational Engineering (AMCCE 2015) © 2015. The authors Published by Atlantis Press 1019 processing, according to the first object of study by wavelet analysis, the wavelet analysis itself has no deformation[2], the wavelet analysis and moment invariants are combined, and the wavelet moment invariant is used to mine the details of the image control ability, in order to obtain the rotation invariant moment, the formula is adopt as: For the continuous grey function ( ) , f x y , the continuous gray function is particularly effective to the moment rotation and the details of the image processing, its ( ) p q + order two dimension origin moment pq M is defined as: ( ) -

Automatic medical image segmentation has wide applications for disease diagnosing. However, it is much more challenging than natural optical image segmentation due to the high-resolution of medical images and the corresponding huge computation cost. The sliding window is a commonly used technique for whole slide image (WSI) segmentation, however, for these methods based on the sliding window, the main drawback is lacking global contextual information for supervision. In this paper, we propose a dual-inputs attention network (denoted as DA-RefineNet) for WSI segmentation, where both local fine-grained information and global coarse information can be efficiently utilized. Sufficient comparative experiments are conducted to evaluate the effectiveness of the proposed method, the results prove that the proposed method can achieve better performance on WSI segmentation compared to methods relying on single-input.

Automatic medical image segmentation has wide applications for disease diagnosing. However, it is much more challenging than natural optical image segmentation due to the high-resolution of medical images and the corresponding huge computation cost. The sliding window is a commonly used technique for whole slide image (WSI) segmentation, however, for these methods based on the sliding window, the main drawback is lacking global contextual information for supervision. In this paper, we propose a dual-inputs attention network (denoted as DA-RefineNet) for WSI segmentation, where both local fine-grained information and global coarse information can be efficiently utilized. Sufficient comparative experiments are conducted to evaluate the effectiveness of the proposed method, the results prove that the proposed method can achieve better performance on WSI segmentation compared to methods relying on single-input.

The liver is an important organ in the human body. As the information provided by 2D images is limited, the 3D liver model is needed to allow doctors to observe the size and shape of the organ more clearly. Most of the traditional 3D modeling methods are based on the difference in CT values of organs. But organs with small differences in CT values cannot be constructed in high quality. To solve this problem, we proposed a 3D modeling method for the liver based on image segmentation. The liver is segmented and then the segmented images are used for 3D modeling, which can effectively reduce the influence of other organs. However, the existing segmentation methods cannot well adapt to the changes in liver shape, size, and position, and the segmentation effect for livers of different shapes is very unsatisfactory. To improve the liver segmentation effect, we added a fully attentional network (FAN) block to the classical U-Net network to enhance the extraction of liver features. Our model achieves 83.56% and 92.53% on MIoU and MPA metrics, respectively, which are 5.51% and 0.69% higher than the classical U-Net.

"Fundamentals of Image Processing" offers a comprehensive exploration of image processing's pivotal techniques, tools, and applications. Beginning with an overview, the book systematically categorizes and explains the multifaceted steps and methodologies inherent to the digital processing of images. The text progresses from basic concepts like sampling and quantization to advanced techniques such as image restoration and feature extraction. Special emphasis is given to algorithms and models crucial to image enhancement, restoration, segmentation, and application. In the initial segments, the intricacies of digital imaging systems, pixel connectivity, color models, and file formats are dissected. Following this, image enhancement techniques, including spatial and frequency domain methods and histogram processing, are elaborated upon. The book then addresses image restoration, discussing degradation models, noise modeling, and blur, and offers insights into the compelling world of multi-resolution analysis with in-depth discussions on wavelets and image pyramids. Segmentation processes, especially edge operators, boundary detections, and thresholding techniques, are detailed in subsequent chapters. The text culminates by diving deep into the applications of image processing, exploring supervised and unsupervised learning, clustering algorithms, and various classifiers. Throughout the discourse, practical examples, real-world applications, and intuitive diagrams are integrated to facilitate an enriched learning experience. This book stands as an essential guide for both novices aiming to grasp the basics and experts looking to hone their knowledge in image processing. Keywords: Digital Imaging Systems, Image Enhancement, Image Restoration, Multi-resolution Analysis, Wavelets, Image Segmentation, Feature Extraction, SIFT, SURF, Image Classifiers, Supervised Learning, Clustering Algorithms.

An efficient approach for structure-texture image decomposition with edge-preservation

In the realm of medical image segmentation, convolutional neural networks (CNN), notably U-Net, have enjoyed considerable success. It was developed for the purpose of biomedical image segmentation, such as identifying and delineating structures or regions of interest within medical images. However, CNN-based techniques grapple with constraints in establishing distant dependencies and global contextual links. While the Transformer's attention mechanism has been proposed to capture remote correlations, it often compromises the precision of local boundary information. In response to this challenge, we introduce a novel two-path network model named C-UVNet, designed for dermoscopy image segmentation. C-UVNet leverages a contextual attention network to surmount these limitations. Our system is constructed on the Visual Transformer (ViT) module, harmonizing it with the classic U-Net architecture to enable a dual focus: highlighting informative regions while accommodating long-range contextual dependencies. Equipped with a robust self-attention mechanism, our encoder adeptly captures both local and global contextual information. This approach empowers our model to excel in segmentation performance, especially in demanding scenarios, as validated using an extensive medical image dataset (ISIC 2017). What’ more, C-UVNet based system surpasses traditional expert accuracy with a range of 75%-84% in medical image segmentation. The results underscore the efficacy of our approach.

In medical image segmentation, convolutional neural networks (CNNs) and transformers are dominant. For CNNs, given the local receptive fields of convolutional layers, long-range spatial correlations are captured through consecutive convolutions and pooling. However, as the computational cost and memory footprint can be prohibitively large, 3D models can only afford fewer layers than 2D models with reduced receptive fields and abstract levels. For transformers, although long-range correlations can be captured by multi-head attention, its quadratic complexity with respect to input size is computationally demanding. Therefore, either model may require input size reduction to allow more filters and layers for better segmentation. Nevertheless, given their discrete nature, models trained with patch-wise training or image downsampling may produce suboptimal results when applied on higher resolutions. To address this issue, here we propose the resolution-robust HNOSeg-XS architecture. We model image segmentation by learnable partial differential equations through the Fourier neural operator which has the zero-shot super-resolution property. By replacing the Fourier transform by the Hartley transform and reformulating the problem in the frequency domain, we created the HNOSeg-XS model, which is resolution robust, fast, memory efficient, and extremely parameter efficient. When tested on the BraTS'23, KiTS'23, and MVSeg'23 datasets with a Tesla V100 GPU, HNOSeg-XS showed its superior resolution robustness with fewer than 34.7k model parameters. It also achieved the overall best inference time (<0.24 s) and memory efficiency (<1.8 GiB) compared to the tested CNN and transformer models. The code repository is available at https://github.com/IBM/multimodal-3d-image-segmentation.

Medical image segmentation is an active research topic to analyse medical images to find an organ or possible abnormalities in an image. Using a Convolutional Neural Network (CNN) is a successful technique for medical image segmentation. However, developing a CNN is a difficult task, especially when it includes complex structures, such as an attention mechanism. A CNN equipped with an attention mechanism is able to focus on a specific part of an image to extract a Region Of Interest (ROI), that can play a significant role to increase the accuracy of an image segmentation. Due to the difficulty of developing an attention network, in this paper, we introduce a new evolutionary technique to generate an attention network automatically for medical image segmentation. To the best of our knowledge, this is the first attempt to create an attention network using an evolutionary technique. To do this, a new encoding model is introduced to create a network topology, along with its training parameters, to ease the complexity of developing a CNN. Also, a Genetic Algorithm (GA) is applied to evolve the networks. To show the capability of the proposed technique, we used three publicly available medical segmentation datasets. The obtained results show that the proposed model can generate networks corresponding to each dataset, such that the developed networks have high performance for medical image segmentation.

Importance of little objects in cars such as door handles is obvious, both in daily lives and in industrial manufacture. However, since the lack of the distinctive appearance and feature, obtaining the location of them is still remaining a challenge. This paper proposes an effective approach for the detection of the door handles of cars. The method innovatively combines frequency and spatial domains' algorithms to detect the location of little objects of cars in a relatively large image. To illustrate the method more concisely, our method localizes door handles through image segmentation and visual saliency detection. First, by segmenting the image we can remove the unnecessary area to improve the speed and accuracy of our approach. After finding the region of interest, our approach uses a visual saliency detection algorithm named Spectral Residual Approach which can get the location of door handles accurately. At last, the approach is tested by different kinds of images of vehicles. The results of the experiments show that our approach is obvious and practical.

Recently, the importance of semantic segmentation research for scene understanding in frontal viewing camera images of autonomous vehicles has increased. The existing state-of-the-art (SOTA) methods for semantic segmentation exhibit high accuracy for high-resolution images and low-resolution (LR) images without degradation factors of blur and noise. Owing to the nature of vehicles, the need is increasing for the pre-judgment of emergencies through the accurate semantic segmentation of LR images with the degradation factors acquired by low-cost camera at far distance. However, no research exists on super-resolution reconstruction (SR)-based semantic segmentation of LR images with degradation factors. Therefore, this study proposes a novel combined network for a super-resolution reconstruction and semantic segmentation (CN4SRSS) framework based on attention and re-focus network (ARNet), which exhibits low computational cost and high semantic segmentation accuracy. The experimental results using LR image datasets based on CamVid and Minicity datasets, which are open databases, show that the semantic segmentation accuracy (pixel accuracy) based on the proposed CN4SRSS and DeepLab v3 + is 93.14% and 89.48%, respectively. Particularly, the proposed method shows higher accuracy when compared to the SOTA methods. Furthermore, the proposed method has been confirmed that requires lower computational cost in terms of the number of parameters, memory usage, number of multi-adds calculation, and floating-point operations per second (FLOPs) than the SOTA methods.

Accurate cerebral vascular endothelium segmentation in Optical Coherence Tomography (OCT) images is crucial for cerebrovascular disease assessment, yet remains challenging due to the extreme thinness of endothelial structures and the scarcity of high-quality annotations. In this work, we make two key contributions. First, we construct a high-quality cerebral vascular OCT dataset with meticulous manual annotations provided by experienced experts, offering a reliable foundation for supervised learning and quantitative evaluation. Second, we propose a novel segmentation framework based on a Dual Coordinate Attention (DCA) mechanism, which explicitly integrates Cartesian and polar coordinate representations to capture complementary structural cues of vascular endothelium. Extensive experiments demonstrate that the proposed DCA-based network consistently outperforms representative baseline models in terms of Dice and HD95. Ablation studies further validate the effectiveness of the DCA module and identify its optimal deployment strategy. Overall, this work provides a robust automated solution for cerebral vascular endothelium segmentation in OCT images, with potential value for cerebrovascular research and clinical assessment.

Accurate medical image segmentation is of utmost importance in a wide range of clinical applications, playing a vital role in disease diagnosis and treatment planning. This research presents the application of the Explainable Multi-Module Semantic Guided Attention Network (EM-SGAN) with the optimization technique of unbounded variance Adaptive Moment Estimation (AMSGrad) for breast cancer image segmentation. EM-SGAN is a deep learning model that integrates multiple modules to enhance the accuracy and interpretability of the segmentation process. The key components of EM-SGAN include an encoder-decoder framework, attention mechanism, semantic guidance module, and explainability module. By incorporating the AMSGrad optimizer, which addresses the unboundedness issue of the second-moment estimate, EM-SGAN achieves stable convergence and improved optimization. Experimental evaluations on breast cancer image segmentation tasks demonstrate the effectiveness of EM-SGAN with unbounded variance AMSGrad in accurately segmenting cancerous regions. The proposed approach significantly advances the field of medical image segmentation by offering a dependable and understandable solution for breast cancer analysis.

Effective hybrid attention network based on pseudo-color enhancement in ultrasound image segmentation

Non-same-scale feature attention network based on BPD for medical image segmentation