Image Clustering Research Articles

An image fusion algorithm based on image clustering theory

An image fusion algorithm based on image clustering theory

Optimized binarization algorithm-based method for the image recognition and characterization of explosion damage in rock masses

The quantitative analysis of rock mass damage is crucial in fields such as engineering geology, disaster prevention, mining, geotechnical engineering, and structural engineering. With the advancement and application of noncontact measurement technologies and fractal theory, image-based damage identification methods are gaining increasing importance. This paper presents an optimized binarization algorithm for identifying and characterizing damage zones in granite explosion images. The method involves filtering, mathematical morphology operations, and connectivity recognition to effectively remove background noise while preserving clear boundaries of the damaged areas. It accurately captures the explosion damage in granite, both in terms of damage morphology and characteristic parameters. Additionally, the coefficient of agreement (COA) is introduced to quantitatively assess the accuracy of different methods in identifying damaged areas. The experimental results show that, compared with commonly used methods such as Otsu's method, Bernsen's algorithm, Niblack's algorithm, Sauvola's algorithm, and the K-means image clustering algorithm, the proposed method performs better in terms of identification accuracy and parameter agreement, achieving COA values near 1 across diverse experimental environments. Furthermore, the proposed method excels in handling uneven lighting, mitigating interference from rock surface textures and explosion carbonization zones, and demonstrates significant robustness in complex scenarios. The findings of this paper provide insights into the integration of engineering geology and computer vision technology. They offer valuable references for damage identification in excavation damage zones (EDZs), geological disaster evaluation, and structural damage warning systems.

Unsupervised contrastive clustering via density cluster representative combination

The recent development of contrastive clustering for deep image clustering has shown promising results by combining representation learning and clustering prediction into a unified framework. In this paper, we introduce an unsupervised contrastive clustering method inspired by contrastive learning to address the issue of cluster center selection in density peaks clustering. We propose a new type of information entropy peaks based on dynamic reverse nearest neighbor sequence to measure local density for each object, and select a small number of top information entropy peaks as cluster representative candidates to cover the right cluster centers as comprehensively as possible. To mitigate the risk of wrongly determining cluster centers, we generate a cluster centers set by computing various cluster representative candidates combinations. In the contrastive clustering stage, we introduce transitive nearest neighbor distance to compute cluster-level loss for the core data set. The prototype clustering is obtained with the minimum cluster-level loss, and the non-core data set is grouped into prototype clustering by nearest neighbor distance. Our comprehensive experiments on both artificial and real-world data sets demonstrate that the proposed unsupervised contrastive clustering method is effective and outperforms the compared state-of-the-art clustering algorithms.

Graph convolutional and random Fourier feature mapping for hyperspectral image clustering

Graph convolutional and random Fourier feature mapping for hyperspectral image clustering

Chemical Landscape of Adipocytes Derived from 3T3-L1 Cells Investigated by Fourier Transform Infrared and Raman Spectroscopies.

Adipocytes derived from 3T3-L1 cells are a gold standard for analyses of adipogenesis processes and the metabolism of fat cells. A widely used histological and immunohistochemical staining and mass spectrometry lipidomics are mainly aimed for examining lipid droplets (LDs). Visualizing other cellular compartments contributing to the cellular machinery requires additional cell culturing for multiple labeling. Here, we present the localization of the intracellular structure of the 3T3-L1-derived adipocytes utilizing vibrational spectromicroscopy, which simultaneously illustrates the cellular compartments and provides chemical composition without extensive sample preparation and in the naïve state. Both vibrational spectra (FTIR-Fourier transform infrared and RS-Raman scattering spectroscopy) extended the gathered chemical information. We proved that both IR and RS spectra provide distinct chemical information about lipid content and their structure. Despite the expected presence of triacylglycerols and cholesteryl esters in lipid droplets, we also estimated the length and unsaturation degree of the fatty acid acyl chains that were congruent with known MS lipidomics of these cells. In addition, the clustering of spectral images revealed that the direct surroundings around LDs attributed to lipid-associated proteins and a high abundance of mitochondria. Finally, by using quantified markers of biomolecules, we showed that the fixative agents, paraformaldehyde and glutaraldehyde, affected the cellular compartment differently. We concluded that PFA preserves LDs better, while GA fixation is better for cytochromes and unsaturated lipid analysis. The proposed analysis of the spectral images constitutes a complementary tool for investigations into the structural and molecular features of fat cells.

A data-driven approach to traditional village colors: K-Means clustering of online images

ABSTRACT This study aims to quantify the color characteristics of traditional villages, explore the key aspects of color that attract tourist attention, and propose solutions to enhance the readability and legibility of these villages. The findings are intended to provide practical guidance for the restoration and use of village colors. The research focuses on how tourists perceive the color characteristics of traditional villages. The objective is to conduct color clustering analysis to identify the composition and characteristics of village colors, and further analyze the underlying causes to offer suggestions for optimization and enhancement. It employs a methodology based on urban imagery theory, using clustering and secondary extraction of color data from online images. The results reveal distinct contrasts and harmonies in the colors of the five elements providing a basis for improving spatial readability and visual guidance. These findings not only offer practical design references for optimizing traditional village colors but also provide valuable insights for the protection and restoration of other traditional sites. In particular, the contrast between the colors of boundaries and landmarks can enhance visual recognition, while the complementary color designs of paths and boundaries can improve spatial readability, contributing to the sustainable development of traditional villages.

Euler Kernel Mapping for Hyperspectral Image Clustering via Self-Paced Learning

Clustering, as a classical unsupervised artificial intelligence technology, is commonly employed for hyperspectral image clustering tasks. However, most existing clustering methods designed for remote sensing tasks aim to solve a non-convex objective function, which can be optimized iteratively, beginning with random initializations. Consequently, during the learning phase of the clustering model, it may easily fall into bad local optimal solutions and finally hurt the clustering performance. Additionally, prevailing approaches often exhibit limitations in capturing the intricate structures inherent in hyperspectral images and are very sensitive to noise and outliers that widely exist in remote sensing data. To address these issues, we proposed a novel Euler kernel mapping for hyperspectral image clustering via self-paced learning (EKM-SPL). EKM-SPL first employs self-paced learning to learn the clustering model in a meaningful order by progressing samples from easy to complex, which can help to remove bad local optimal solutions. Secondly, a probabilistic soft weighting scheme is employed to measure complexity across the data sample, which makes the optimization process more reasonable. Thirdly, in order to more accurately characterize the intricate structure of hyperspectral images, Euler kernel mapping is used to convert the original data into a reproduced kernel Hilbert space, where the nonlinearly inseparable clusters may become linearly separable. Moreover, we innovatively integrate the coordinate descent technique into the optimization algorithm to circumvent the computational inefficiencies and information loss typically associated with conventional kernel methods. Extensive experiments conducted on classic benchmark hyperspectral image datasets illustrate the effectiveness and superiority of our proposed model.

A Comparative Study of Mean Square Error, Dimensions, Signal to Noise Ratio of Colored and Non-Colored Clustered Original Images Along with Compressed Version After the Image Segmentation and Filtering Method

Primarily author has already done one fundamental paper work on image clustering and segmentation but here in this paper author has continued that same type of work on clustered and segmented images as a mode of comparative study for author has chosen three different parameters like mean square error, peak SNR and dimensions of images (length, width, height). The author has all three parametric methods on one particular to justify the comparison. So this paper is a cumulative case of a comparative study for which author has chosen the above mentioned parameters to justify the best results of the clustered and segmented images.

Revisiting PD: Unraveling the brain-first and body-first perspectives

Parkinson’s disease (PD) has undergone significant advancements in diagnosis and treatment over the past century, with apparent dopaminergic cell degeneration on dopamine transporter scans and a strong response to medication being key features. However, the etiology remains complex, involving various pathogenic mechanisms beyond alpha-synuclein accumulation. The recent brain-first versus body-first hypothesis, emerging from advances in functional imaging and clinical symptom clustering, suggests distinct starting points of alpha-synuclein pathology-either in the brain or the body, with subsequent spread via neural connections. This theory, exemplified by the alpha-synuclein origin site and connectome (SOC) model, proposes that body-first PD may originate in the enteric nervous system and spread to the brain, while brain-first PD starts within the central nervous system, such as the olfactory bulb or amygdala. While the SOC model offers valuable insights into the progression of PD, it raises several controversies. Critics argue that the model may oversimplify the disease’s complexity, failing to account for overlapping symptoms and the varying progression rates observed in different subtypes. Furthermore, there are concerns about the lack of longitudinal data and the potential for reclassification of PD subtypes over time. Despite these challenges, the ongoing development of imaging techniques that reflect in-vivo pathology holds promise for resolving these controversies and advancing the selection of patients for disease-modifying therapies.

Self-supervised based clustering for retinal optical coherence tomography images.

In response to the inadequacy of manual analysis in meeting the rising demand for retinal optical coherence tomography (OCT) images, a self-supervised learning-based clustering model was implemented. A public dataset was utilized, with 83,484 OCT images with categories of choroidal neovascularization (CNV), diabetic macular edema (DME), drusen, and normal fundus. This study employed the Semantic Pseudo Labeling for Image Clustering (SPICE) framework, a self-supervised learning-based method, to cluster unlabeled OCT images into binary and four categories, and the performances were compared with baseline models. We also analysed feature distribution using t-SNE, and explored the cluster centers, attention maps, and misclassified images. In addition, DME and CNV subsets were clustered binarily, and the results were interpreted by two retinal specialists. SPICE demonstrated superior performance in binary and four categories classification tasks, achieving the accuracy of 0.886 and 0.846, respectively. In t-SNE analysis, the four types exhibited significant clustering into distinct groups. The cluster centers corresponded to the human labels, and the heat map revealed that the model focused on important biomarkers. The misclassified images exposed similar features to the inaccurate classes. The model also grouped DME and CNV into two distinct categories respectively. Self-supervised clustering effectively distinguished disease variances and revealed common features, with a notable capability to detect disease heterogeneity through biomarkers.

Distributed Fire Classification and Localization Model Based on Federated Learning with Image Clustering

Distributed Fire Classification and Localization Model Based on Federated Learning with Image Clustering

A geometric view of seismic wavefields: implications for imaging dense clusters of events

SUMMARY Imaging dense clusters of seismicity is crucial to many problems in seismology: to delineate complex systems of faults, provide constraints on the causes of volcanic and cryogenic swarms, and to shed light on possible means to prevent damaging induced seismicity in mining, geothermal, and oil and gas extraction activities. Current imaging methods rely upon high-resolution relative location techniques, commonly requiring arrival-time picks for seismic phases. This paper examines an alternative approach, based upon concepts drawn from differential geometry, that images directly from waveform data. It relies upon the common assumption of spatial continuity of seismic wavefield observations, which implies that a differentiable map exists between the source region to be imaged and waveform observations considered as elements of a vector space. The map creates an image of event clusters on a Riemannian manifold embedded in that vector space. The image can be visualized by projecting the observations into a tangent space of the manifold and is a distorted rendering of cluster geometry. However, the distortion can be predicted and removed if a model for wavefield propagation is available. This visualization approach is applicable to clusters of uniform events with highly similar waveforms, such as are commonly acquired with correlation detectors or other pattern matching techniques. To assess its performance, it is applied to the closely related reciprocal problem of imaging the (known) geometry of an array from observations by the array of several regional events. Differences between the original problem and its reciprocal analogue are noted and controlled for in the analysis. Chief among the differences is the necessity for aligning the waveforms in the original problem, which, to maintain consistency with the original problem, is solved in the reciprocal problem by a generalization of the VanDecar–Crosson algorithm. The VanDecar–Crosson algorithm exhibits a bias, shown through an analysis of the situation when the observed wavefields are adequately modelled as plane waves. In that circumstance, the bias can be predicted and removed. In a test using a portion of a large-N array, this imaging approach is shown to successfully reconstruct the array geometry. The method is applicable directly to infinitesimal array apertures, but is extended to a larger aperture by partitioning the image into local, effectively infinitesimal overlapping subsets. These are inverted, then assembled into a global picture of the array geometry using constraints provided by the overlapped regions. Although demonstrated in a reciprocal array context, the method appears viable for imaging clusters of events with highly similar source mechanisms and time histories.

Deep learning to capture leaf shape in plant images: Validation by geometric morphometrics.

Plant leaves play a pivotal role in automated species identification using deep learning (DL). However, achieving reproducible capture of leaf variation remains challenging due to the inherent "black box" problem of DL models. To evaluate the effectiveness of DL in capturing leaf shape, we used geometric morphometrics (GM), an emerging component of eXplainable Artificial Intelligence (XAI) toolkits. We photographed Ranunculus auricomus leaves directly in situ and after herbarization. From these corresponding leaf images, we automatically extracted DL features using a neural network and digitized leaf shapes using GM. The association between the extracted DL features and GM shapes was then evaluated using dimension reduction and covariation models. DL features facilitated the clustering of leaf images by source populations in both in situ and herbarized leaf image datasets, and certain DL features were significantly associated with biological leaf shape variation as inferred by GM. DL features also enabled leaf classification into morpho-phylogenomic groups within the intricate R. auricomus species complex. We demonstrated that simple in situ leaf imaging and DL reproducibly captured leaf shape variation at the population level, while combining this approach with GM provided key insights into the shape information extracted from images by computer vision, a necessary prerequisite for reliable automated plant phenotyping.

Efficient algorithms of box-constrained Nonnegative Matrix Factorization and its applications in image clustering

Nonnegative Matrix Factorization (NMF) provides an important approach of unsupervising learning, but it faces computational challenges when applied into clustering of high-dimensional datasets. In this paper, a class of novel nonmonotone gradient-descent algorithms are developed for solving box-constrained NMF problems. Unlike existing algorithms in the literature that update each matrix factor individually by fixing another one, our algorithms simultaneously update the paired matrix factors by leveraging adaptive projected Barzilai-Borwein directions and appropriate step sizes generated by the developed nonmonotone line search rules. Theoretically, it is proved that the developed algorithms are well defined and globally convergent. Extensive numerical tests on public image datasets demonstrate that the developed algorithms in this paper outperform the state-of-the-art ones, in terms of clustering performance, computational efficiency, and robustness of mining noisy data.

Parameterization before Meta-Analysis: Cross-Modal Embedding Clustering for Forest Ecology Question-Answering

In the field of forestry ecology, image data capture factual information, while literature is rich with expert knowledge. The corpus within the literature can provide expert-level annotations for images, and the visual information within images naturally serves as a clustering center for the textual corpus. However, both image data and literature represent large and rapidly growing, unstructured datasets of heterogeneous modalities. To address this challenge, we propose cross-modal embedding clustering, a method that parameterizes these datasets using a deep learning model with relatively few annotated samples. This approach offers a means to retrieve relevant factual information and expert knowledge from the database of images and literature through a question-answering mechanism. Specifically, we align images and literature across modalities using a pair of encoders, followed by cross-modal information fusion, and feed these data into an autoregressive generative language model for question-answering with user feedback. Experiments demonstrate that this cross-modal clustering method enhances the performance of image recognition, cross-modal retrieval, and cross-modal question-answering models. Our method achieves superior performance on standardized tasks in public datasets for image recognition, cross-modal retrieval, and cross-modal question-answering, notably achieving a 21.94% improvement in performance on the cross-modal question-answering task of the ScienceQA dataset, thereby validating the efficacy of our approach. Essentially, our method targets cross-modal information fusion, combining perspectives from multiple tasks and utilizing cross-modal representation clustering of images and text. This approach effectively addresses the interdisciplinary complexity of forestry ecology literature and the parameterization of unstructured heterogeneous data encapsulating species diversity in conservation images. Building on this foundation, intelligent methods are employed to leverage large-scale data, providing an intelligent research assistant tool for conducting forestry ecological studies on larger temporal and spatial scales.

Double-level deep multi-view collaborative learning for image clustering

Double-level deep multi-view collaborative learning for image clustering

SimMolCC: A Similarity of Automatically Detected Bio-Molecule Clusters between Fluorescent Cells

In the field of studies on the “Neural Synapses” in the nervous system, its experts manually (or pseudo-automatically) detect the bio-molecule clusters (e.g., of proteins) in many TIRF (Total Internal Reflection Fluorescence) images of a fluorescent cell and analyze their static/dynamic behaviors. This paper proposes a novel method for the automatic detection of the bio-molecule clusters in a TIRF image of a fluorescent cell and conducts several experiments on its performance, e.g., mAP @ IoU (mean Average Precision @ Intersection over Union) and F1-score @ IoU, as an objective/quantitative means of evaluation. As a result, the best of the proposed methods achieved 0.695 as its mAP @ IoU = 0.5 and 0.250 as its F1-score @ IoU = 0.5 and would have to be improved, especially with respect to its recall @ IoU. But, the proposed method could automatically detect bio-molecule clusters that are not only circular and not always uniform in size, and it can output various histograms and heatmaps for novel deeper analyses of the automatically detected bio-molecule clusters, while the particles detected by the Mosaic Particle Tracker 2D/3D, which is one of the most conventional methods for experts, can be only circular and uniform in size. In addition, this paper defines and validates a novel similarity of automatically detected bio-molecule clusters between fluorescent cells, i.e., SimMolCC, and also shows some examples of SimMolCC-based applications.

Learning clustering-friendly representations via partial information discrimination and cross-level interaction

Despite significant advances in the deep clustering research, there remain three critical limitations to most of the existing approaches. First, they often derive the clustering result by associating some distribution-based loss to specific network layers, neglecting the potential benefits of leveraging the contrastive sample-wise relationships. Second, they frequently focus on representation learning at the full-image scale, overlooking the discriminative information latent in partial image regions. Third, although some prior studies perform the learning process at multiple levels, they mostly lack the ability to exploit the interaction between different learning levels. To overcome these limitations, this paper presents a novel deep image clustering approach via Partial Information discrimination and Cross-level Interaction (PICI). Specifically, we utilize a Transformer encoder as the backbone, coupled with two types of augmentations to formulate two parallel views. The augmented samples, integrated with masked patches, are processed through the Transformer encoder to produce the class tokens. Subsequently, three partial information learning modules are jointly enforced, namely, the partial information self-discrimination (PISD) module for masked image reconstruction, the partial information contrastive discrimination (PICD) module for the simultaneous instance- and cluster-level contrastive learning, and the cross-level interaction (CLI) module to ensure the consistency across different learning levels. Through this unified formulation, our PICI approach for the first time, to our knowledge, bridges the gap between the masked image modeling and the deep contrastive clustering, offering a novel pathway for enhanced representation learning and clustering. Experimental results across six image datasets demonstrate the superiority of our PICI approach over the state-of-the-art. In particular, our approach achieves an ACC of 0.772 (0.634) on the RSOD (UC-Merced) dataset, which shows an improvement of 29.7% (24.8%) over the best baseline. The source code is available at https://github.com/Regan-Zhang/PICI.

ECCT: Efficient Contrastive Clustering via Pseudo-Siamese Vision Transformer and Multi-view Augmentation

Image clustering aims to divide a set of unlabeled images into multiple clusters. Recently, clustering methods based on contrastive learning have attracted much attention due to their ability to learn discriminative feature representations. Nevertheless, existing clustering algorithms face challenges in capturing global information and preserving semantic continuity. Additionally, these methods often exhibit relatively singular feature distributions, limiting the full potential of contrastive learning in clustering. These problems can have a negative impact on the performance of image clustering. To address the above problems, we propose a deep clustering framework termed Efficient Contrastive Clustering via Pseudo-Siamese Vision Transformer and Multi-view Augmentation (ECCT). The core idea is to introduce Vision Transformer (ViT) to provide the global view, and improve it with Hilbert Patch Embedding (HPE) module to construct a new ViT branch. Finally, we fuse the features extracted from the two ViT branches to obtain both global view and semantic coherence. In addition, we employ multi-view random aggressive augmentation to broaden the feature distribution, enabling the model to learn more comprehensive and richer contrastive features. Our results on five datasets demonstrate that ECCT outperforms previous clustering methods. In particular, the ARI metric of ECCT on the STL-10 (ImageNet-Dogs) dataset is 0.852 (0.424), which is 10.3% (4.8%) higher than the best baseline.

Quantifying Surface Shelf Water Export in the Southern Middle Atlantic Bight Using a Lagrangian Particle Tracking Approach

AbstractShelf water is influenced by atmospheric forcing, river outflows, and the open ocean. Studying its variability is crucial for understanding anthropogenic impacts on coastal oceans and their transport to the open ocean. In the Middle Atlantic Bight (MAB), the interaction of the Gulf Stream with shelf/slope circulation leads to some of the complex exchanges between the shelf and open ocean along the U.S. East Coast. This study employs a Lagrangian particle tracking approach, grounded in a high‐resolution, data‐assimilative ocean reanalysis, to examine the export pathways of surface shelf water in the MAB. We analyzed over 700 daily images of simulated particle distributions using image clustering techniques. This revealed three distinct export patterns: abrupt entrainment to the Gulf Stream, gradual entrainment, and southern transport. Each pattern was observed roughly equally during the study period from January 2017 to December 2018. The observed export patterns are closely linked to the coastal circulation dynamics near Cape Hatteras. Understanding the timing and duration of these patterns is vital for assessing water quality and predicting the settlement of species that spawn in the region. Our study further underscores the influence of tropical cyclones, including Hurricanes Jose, Maria, and Chris, on these export patterns. These extreme weather events lead to significant shifts in coastal circulation near Cape Hatteras.