China
Africa
Explore chapters and articles related to this topic
Basic Approaches of Artificial Intelligence and Machine Learning in Thermal Image Processing
Published in U. Snekhalatha, K. Palani Thanaraj, Kurt Ammer, Artificial Intelligence-Based Infrared Thermal Image Processing and Its Applications, 2023
U. Snekhalatha, K. Palani Thanaraj, Kurt Ammer
Cluster-based image segmentation involves the grouping of the image pixels into groups also known as clusters (Mittal et al., 2021). The pixels in each group are as similar to each other as possible, while pixels belonging to different groups are as dissimilar to each other as possible. Also, the distance between cluster centers is also kept as far as possible. The common principle behind the different clustering algorithms is that initially, the number of clusters is to be specified by the user. Next, the centroid is calculated, and the pixels are allotted to each cluster accordingly. More iterations are performed to the centroid calculation until it no longer changes, i.e., the algorithm converges. Some of the important and frequently used algorithms in medical image processing are explained as follows.
Cancerous or Non-Cancerous Cell Detection on a Field-Programmable Gate Array Medical Image Segmentation Using Xilinx System
Published in Neeraj Mohan, Surbhi Gupta, Chuan-Ming Liu, Society 5.0 and the Future of Emerging Computational Technologies, 2022
C. Gopala Krishnan, Prasannavenkatesan Theerthagiri, A.H. Nishan
Without significant efforts during data preparation, these data cannot be used for data mining, which includes two techniques: image registration and image segmentation (Gupta and Pahuja, 2017). Image segmentation is a method of separating data into a continuous region for representing individual anatomical objects. It is used for many applications in the computer-related medical field in which they can classify different tissues in to separate logical classes. The major problem in image segmentation is about how to separate interested object from background noise. These different forms help to derive information that helps for more data-mining tasks. The amount of rotation and the translation of images in any direction helps to distinguish medical images, and they may also differ in their scale (Prasannavenkatesan, 2021).
Image Segmentation
Published in Rashmi Gupta, Arun Kumar Rana, Sachin Dhawan, Korhan Cengiz, Advanced Sensing in Image Processing and IoT, 2022
Hina J. Chokshi, Abhishek Agarwal
Neural networks play an important role in image segmentation. Artificial intelligence is used to examine an image and identify the objects, human faces, text data, and other required information. Convolutional neural networks (CNNs) are said to be preferred for image segmentation, as they can extract the different image features and statistics at a much faster rate and with highest accuracy [36]. Recently, the experts at Facebook AI analysis (FAIR) created a Mask R-CNN, which is a deep gaining knowledge of layout that created pixel-clever masks for individual image elements. It falls into the category of an improved model of the faster R-CNN item detection architecture. For each element present in the image, the quicker R-CNN employs two distinct fact devices. Other additional information associated with an image can be extracted using the Mask R-CNN algorithm. This algorithm first generates the function map of the image. Then, the device applies the area idea network (RPN) at the function maps and generates the element proposals with their objectless scores. As soon as it is completed, the pooling layer is applied to the suggestions, allowing them to be conveyed in a uniform manner [37]. Finally, the proposals to the corresponding layer for class are passed by the device, and the output is generated.
Recognition of PCM in soil using semantic segmentation model and numerical simulation
Published in Particulate Science and Technology, 2023
Weijie Mao, Biao Ma, Qianqi Zhang, Xu Hua
The primary problem to study the mesoscopic heat transfer phenomenon of soil containing PCM is to obtain its geometric model. Microscope technology is widely used to observe the microscopic morphology of materials, which produces realistic geometric images of materials at microscopic scale (Zhou et al. 2021; Essid, Eddhahak, and Neji 2022). Typically, these images are used as the primary input stream for various computational vision techniques to quantify the distribution of materials (Paneru and Jeelani 2021; Sun and Gu 2022). For mixture images containing only PCM and soil, a template of the geometric model can be obtained by accurately dividing them into different areas in the image. Traditional image segmentation methods include threshold-based segmentation method, edge-based segmentation method, region-based segmentation method and energy-function-based segmentation method. Specifically, they include OTSU algorithm, superpixel algorithm, watered segmentation algorithm and active contour algorithm, etc. (Ciecholewski 2015; Goh et al. 2018; Stutz, Hermans, and Leibe 2018; Chen et al. 2019). These methods have defects such as poor anti-noise performance and over-segmentation. Deep learning is a data-driven approach that many studies have applied to inspection projects in road engineering (Yang et al. 2021; Li et al. 2022; Liu et al. 2022; Xu and Liu 2022). Among them, semantic segmentation and object detection are the two mainstream methods. Semantic segmentation is better for qualitative and quantitative analysis than object detection (Nan et al. 2021).
An advanced fuzzy C-Means algorithm for the tissue segmentation from brain magnetic resonance images in the presence of noise and intensity inhomogeneity
Published in The Imaging Science Journal, 2023
Sandhya Gudise, K. Giri Babu, T. Satya Savithri
Clustering-based segmentation methods effectively segment the medical images even in the presence of IIH. Clustering is an example of an unsupervised image segmentation technique in which the image gets divided into groups of homogeneous pixels based on some prescribed criterion. Fuzzy C-Means (FCM) is a popular clustering-based method. FCM techniques are widely used for brain MRI segmentation in which three clusters for WM, GM, and CSF are estimated by iteratively finding the mean intensity of each tissue class [14–16]. FCM partially segments the voxels into various tissue classes based on the membership functions of every tissue type. These fuzzy clustering approaches have several advantages. In these techniques, each pixel can belong to more than one class which gives the ability to eliminate PVE [17]. It can segment different tissues simultaneously which is a superior performance to some popular techniques like deformable models and level sets [18,19], which can segment only one class of tissue at any time. The main limitation of FCM is its non-convergence to optimal solution owing to the random initialization of cluster centroids. FCM performance is decided by the initialization of cluster centroids. Hence, centroid selection is very important. It also suffers from initialization sensitivity and is very sensitive to artefacts such as noise and bias field [20]. Hence Bias field correction is required before brain MRI segmentation to avoid the wrong estimation of initial cluster centres by intensity normalization. Different research works are described in the literature to solve the shortcomings of FCM [21,22].
Multi-scale convolution based breast cancer image segmentation with attention mechanism in conjunction with war search optimization
Published in International Journal of Computers and Applications, 2023
B. N. Madhukar, S. H. Bharathi, Ashwin M. Polnaya
Breast tumor segmentation must be done accurately to diagnose breast cancer, a treatment planning, and assess the effectiveness of that treatment, [29]. Since manually segmenting breast tumors requires a lot of work, significant effort has gone into developing semi-automated or automatic tumor segmentation techniques. Image segmentation research has shown success using deep learning approaches, mostly CNNs. There are typically millions or even billions of criteria in a deep learning CNN model. Thence, it will offer better improvements in segmenting the image. The proposed breast cancer segmentation technique contains four stages such as pre-processing, augmentation, segmenting image using multi-scale convolution, and multi-attention mechanisms, respectively. To make the proposed system, the breast cancer dataset is used. Initially, Feature wise data augmentation is performed, which allows better training of the deep neural networks. Through various processing techniques or combinations of multiple processing, such as random rotation, shear, shifts, flips, etc., image augmentation artificially generates multiple training images [9]. Then, image resizing is done using bi-cubic interpolation. The noise in the images is denoised by using Pyramid mean Shift filtering. A novel multi-scale convolution and triple attention mechanism as gated axial, position and channel attention based deep learning framework is proposed for tumor image segmentation. Additionally, WSO Algorithm is adapted for fine turning the proposed network. Figure 1 demonstrates the proposed methodology.