Original Image Research Articles

BUILT-IN LENS CORRECTION PROFILES IN LOW-COST CAMERAS: AN ISSUE FOR PHOTOGRAMMETRIC APPLICATIONS?

Abstract. Photogrammetric applications nowadays envisage the use of more and more low-cost cameras such as those equipped on commercial UAV platforms. Typically, these low-grade cameras suffer from extreme radial distortion and strong vignetting among other defects. This, initiated a trend among the low-cost cameras’ manufacturers to try to hide the camera defects by applying software pre-corrections to the images. These Built-In Correction Profiles gets applied to both the JPG files, directly in-camera, and usually to the raw files as well, through the opcode functions of the DNG standard. In this paper we rise this issue that is still under-reported in the literature and further assess the accuracy implication of applying or discarding the Built-In Correction Profile in the scenario of UAV mapping. We tested the commercial UAV DJI Phantom 4 Pro v2 in a calibration environment and a field test to compare the performance of pre-corrected versus uncorrected images. In our tests, processing the original uncorrected images led to improved IO calibration and reduced bowing effect in the field test.

FLATTENING COMPLEX ARCHITECTURAL SURFACES: PHOTOGRAMMETRIC 3D MODELS CONVERTED INTO 2D MAPS

Abstract. The paper describes a workflow to flatten 3D photogrammetric models of undevelopable surfaces into unfragmented 2D texture maps. The aim is to create a texture map with reduced fragmentation compared to typical photogrammetric texture files associated with 3D models. Geometric reformatting of the mesh is required to achieve an unfragmented final texture image with enough visual quality to allow for its use in 2D editing software (e.g., Photoshop, Illustrator, etc.). With a lighter model and defragmented texture, graphic documentation of conditions, treatments, or other relevant information can be performed directly. The approach considers both simple and complex architectural surfaces, with particular attention to elements that cannot be developed without introducing distortions. In the case of historic buildings, such surfaces constitute the majority, especially in the case of decorative elements and irregular vaulted systems. The approach extends to full 3D models, particularly those where orthomosaics would result in stretched details. We will discuss two methods: the first is particularly suitable when the photogrammetric project (oriented images) is still available, and the second applies to generic 3D models without the availability of the original images.

Deep convolutional-neural-network-based metal artifact reduction for CT-guided interventional oncology procedures (MARIO).

Computed tomography (CT) is routinely used to guide cryoablation procedures. Notably, CT-guidance provides 3D localization of cryoprobes and can be used to delineate frozen tissue during ablation. However, metal-induced artifacts from ablation probes can make accurate probe placement challenging and degrade the ice ball conspicuity, which in combination could lead to undertreatment of potentially curable lesions. In this work, we propose an image-based neural network (CNN) model for metal artifact reduction for CT-guided interventional procedures. An image domain metal artifact simulation framework was developed and validated for deep-learning-based metal artifact reduction for interventional oncology (MARIO). CT scans were acquired for 19 different cryoablation probe configurations. The probe configurations varied in the number of probes and the relative orientations. A combination of intensity thresholding and masking based on maximum intensity projections (MIPs) was used to segment both the probes only and probes + artifact in each phantom image. Each of the probe and probe + artifact images were then inserted into 19 unique patient exams, in the image domain, to simulate metal artifact appearance for CT-guided interventional oncology procedures. The resulting 361 pairs of simulated image volumes were partitioned into disjoint training and test datasets of 304 and 57 volumes, respectively. From the training partition, 116600 image patches with a shape of 128×128×5 pixels were randomly extracted to be used for training data. The input images consisted of a superposition of the patient and probe + artifact images. The target images consisted of a superposition of the patient and probe only images. This dataset was used to optimize a U-Net type model. The trained model was then applied to 50 independent, previously unseen CT images obtained during renal cryoablations. Three board-certified radiologists with experience in CT-guided ablations performed a blinded review of the MARIO images. A total of 100 images (50 original, 50 MARIO processed) were assessed across different aspects of image quality on a 4-point likert-type item. Statistical analyses were performed using Wilcoxon signed-rank test for paired samples. Reader scores were significantly higher for MARIO processed images compared to the original images across all metrics (all p<0.001). The average scores of the overall image quality, iceball conspicuity, overall metal artifact, needle tip visualization, target region confidence, and worst metal artifact, needle tip visualization, iceball conspicuity, and target region confidence improved by 34.91%, 36.29%, 39.94%, 34.17%, 35.13%, and 45.70%, respectively. The proposed method of image-based metal artifact simulation can be used to train a MARIO algorithm to effectively reduce probe-related metal artifacts in CT-guided cryoablation procedures.

DNNAM: Image inpainting algorithm via deep neural networks and attention mechanism

Most image inpainting algorithms have problems such as fuzzy images, texture distortion and semantic inaccuracy, and the image inpainting effect is limited when processing photos with large missing sections and resolution levels. The paper proposes an effective image inpainting algorithm via partial multi-scale channel attention mechanism and deep neural networks to address the above phenomenon that existing image inpainting methods using deep learning modules have insufficient perception and representation capabilities for multi-scale features with high proportion of irregular defects. Initially, we used the Res-U-Net module as a generator. The U-Net-like backbone network topology can achieve the encoding and decoding stages of damaged images. Secondly, the residual network structure was built in the encoder and decoder to improve the ability of the proposed network to extract and display the features of the damaged images. Finally, the partial multi-scale channel attention module was inserted in the skip connection with the decoder to increase the efficiency of using the low-level features of the original images. The experimental results of the research can show that the proposed method outperforms state-of-the-art methods in terms of subjective visual perception and objective evaluation indicators on the CelebA, Places2 and Paris Street View datasets.

Detection of Microdefects in Fabric with Multifarious Patterns and Colors Using Deep Convolutional Neural Network

Automatic detection of fabric defects is important in textile quality control, particularly in detecting fabrics with multifarious patterns and colors. This study proposes a fabric defect detection system for fabrics with complex patterns and colors. The proposed system comprises five convolutional layers designed to extract features from the original images effectively. In addition, three fully connected layers are designed to classify the fabric defects into four categories. Using this system, the detection accuracy is improved, and the depth of the model is shortened simultaneously. Optimal detection rates for testing dirty marks, clip marks, broken yams, and defect-free were 88.01%, 90.15%, 98.01%, and 97.73%, respectively. The experimental results show that the proposed method is effective, feasible, and has significant potential for fabric defect detection.

Verification of image quality improvement of low-count bone scintigraphy using deep learning.

To improve image quality for low-count bone scintigraphy using deep learning and evaluate their clinical applicability. Six hundred patients (training, 500; validation, 50; evaluation, 50) were included in this study. Low-count original images (75%, 50%, 25%, 10%, and 5% counts) were generated from reference images (100% counts) using Poisson resampling. Output (DL-filtered) images were obtained after training with U-Net using reference images as teacher data. Gaussian-filtered images were generated for comparison. Peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) to the reference image were calculated to determine image quality. Artificial neural network (ANN) value, bone scan index (BSI), and number of hotspots (Hs) were computed using BONENAVI analysis to assess diagnostic performance. Accuracy of bone metastasis detection and area under the curve (AUC) were calculated. PSNR and SSIM for DL-filtered images were highest in all count percentages. BONENAVI analysis values for DL-filtered images did not differ significantly, regardless of the presence or absence of bone metastases. BONENAVI analysis values for original and Gaussian-filtered images differed significantly at ≦25% counts in patients without bone metastases. In patients with bone metastases, BSI and Hs for original and Gaussian-filtered images differed significantly at ≦10% counts, whereas ANN values did not. The accuracy of bone metastasis detection was highest for DL-filtered images in all count percentages; the AUC did not differ significantly. The deep learning method improved image quality and bone metastasis detection accuracy for low-count bone scintigraphy, suggesting its clinical applicability.

Using photo editing to understand the impact of species aesthetics on support for conservation

Abstract Many threatened species suffer from a lack of conservation attention compared to others. Prioritisation of funding, research and conservation efforts seem to be driven by reasons beyond conservation need. This could be due to a ‘beauty bias’, whereby aesthetically pleasing species receive more attention. We examined how editing an image to increase a species' aesthetic appeal may impact donation choices and public attitude towards that species. We posed two research questions; first, ‘do people make different donation choices when they see original images of a species compared to when they see images of the same species that have been edited to match aesthetic preferences?’ Using hypothetical donation experiments, we asked respondents to allocate funds to the conservation of three pictured species, one ‘aesthetically appealing’, one ‘aesthetically unappealing’, and one whose image was either edited to reflect common aesthetic preferences or left unedited. Our findings suggest that images edited to make an animal more visually appealing tend to receive higher hypothetical donation amounts than original images. We also posed a second research question; ‘How do people of varying conservation expertise respond to original versus edited images of wildlife?’ To investigate this, we ran three focus groups with individuals unfamiliar with our test species, those familiar with two or more of our test species, and with conservation professionals, which showed mixed reactions both within and between groups. Focus group participants with less conservation expertise noted that edited images often seemed ‘cuter’ than unedited images, and were more likely to compare them to cartoon characters. Participants with more conservation expertise and species familiarity reported greater empathy towards unedited images, and noted that the edited images prompted an ‘uncanny valley’ response, highlighting the need for further scrutiny in how photo editing might be used in conservation messaging. Our findings support the beauty bias hypothesis and highlight that decisions on conservation support should acknowledge that less aesthetically pleasing species are disadvantaged in public attention and funding. In addition, the findings highlight the role of conservation expertise in impacting viewer reactions, as well as the ethical implications of editing images of wildlife. Read the free Plain Language Summary for this article on the Journal blog.

The debiased morphological transformations of galaxies since z=3 in CANDELS

Morphological quantitative measurements and visual-like classifications are susceptible to biases arising from the expansion of the Universe. One of these biases is the effect of cosmological surface brightness dimming (CSBD): the measured surface brightness of a galaxy decays with redshift as $(1+z)^ $. This effect might lead an observer to perceive an altered morphology compared to the real one. Our goal is to investigate the impact of CSBD on morphological classifications to determine the true evolution of morphological classes over redshift for field galaxies, and to interpret these results in the context of morphological transformations and star formation quenching. We employed artificial redshifting techniques on a sample of 268 galaxies in the five CANDELS fields, spanning redshifts from $z=0.2$ to $z=3.0$. We compared the visual classifications and morphological coefficients ($G$, $M_ $, and $A_s$) obtained from the original and simulated images. Subsequently, we developed two correction methods to mitigate the effects of CSBD. Our findings reveal that CSBD, low resolution, and signal-to-noise significantly bias the visual morphological classifications beyond $z&gt;1$. Specifically, we observed an overestimation of the fractions of spheroids and irregular galaxies by up to $50&lt;!PCT!&gt;$, while the fractions of early- and late-type disks were underestimated by $10&lt;!PCT!&gt;$ and $50&lt;!PCT!&gt;$, respectively. However, we found that morphological coefficients are not significantly affected by CSBD at $z&lt;2.25$. We validated the consistency of our correction methods by applying them to the observed morphological fractions in the IllustrisTNG-50 sample and comparing them to previous studies. We propose two potential sources of confusion regarding the visual classifications due to CSBD. Firstly, galaxies may be misclassified as spheroids, as the dimming effect primarily renders the bulge component visible. Secondly, galaxies may be misidentified as irregulars due to their more diffuse and asymmetric appearance at high redshifts. By analyzing the morphological fractions of star-forming and quiescent subsamples as a function of redshift and stellar mass, we propose a scenario where late-type disks transform into quiescent spheroids through mergers or to early-type disks through secular evolution or active galactic nucleus feedback.

Enhancing Facial Recognition in Surveillance Systems through Embedded Super-resolution

This document details the implementation of a sub-pixel convolutional neural network designed to enhance the resolution of face images. The model uses a series of filters to progressively increase the number of pixels, estimating the necessary information for new pixels from the original image and training derived from 22000 synthetic images produced by adversarial neural networks. Within the context of surveillance and related applications, the trained convolutional network exhibits beneficial characteristics. For instance, it can be deployed within a device to achieve higher-resolution images than those the physical camera can produce. This research underscores the feasibility of such a device through the implementation and evaluation of the network on the NVIDIA Jetson TX2 embedded system. The findings demonstrate the model's practicality for real-time surveillance applications and its ability to produce superior-quality images compared to several interpolation methods, as determined by an exhaustive testing process measuring various attributes of the generated images.

Advancing Cloud Image Security via AES Algorithm Enhancement Techniques

Communication system and internet dominance in our society, has made image security a matter of paramount concern. Cryptography involves encrypting data to protect information exchange between senders and receivers, establishing a foundation for secure communication. The Advanced Encryption Standard (AES) is an exceptional algorithm that plays a pivotal role in this area because of its ability to consistently transform plain data into cipher data using the same encryption key. This algorithm engages intricate encryption techniques, harnessing a variety of algorithms and transformations to ensure robust data security. This study introduces an image encryption technique to comprehensively address security requirements. The proposed method uses the AES algorithm to provide high reliability and unparalleled safety, effectively protecting precious images against any form of malicious attack. The sender's image is intelligently fragmented into multiple pieces to maximize data confidentiality, employing a sophisticated algorithm. Upon decryption by the intended recipient, these fragments seamlessly merge to reconstruct the original image. The suggested technique allows users to securely store their data in the cloud, mitigating concerns related to potential data leakage, damage, or corruption. Thus, by integrating a cloud-based storage system, and utilizing digital signatures through OpenSSL, this method guarantees airtight security and reliability for sensitive data.

Speckle Noise Removal from Biomedical MRI Images and Classification by Multi-Support Vector Machine

INTRODUCTION: Image Processing (IP) methods play a vital role in medical images for diagnosing and predicting illness, as well as monitoring the patient's progress. The IP methods are utilized in many applications for example in the field of medicine. OBJECTIVES: The images that are obtained by the MRI magnetic Resonance imaging and x rays are analyzed with the help of image processing. METHODS: This application is very costly to the patient. Because of the several non-idealities in the image process, medical images are frequently tainted by impulsive, multiplicative, and addictive noise. RESULTS: By replacing some of the original image's pixels with new ones that have luminance values which are less than the allowed dynamic luminance range, noise frequently affects medical images. CONCLUSION: In this research work, the Speckle type noises are eliminated with the help of Mean Filter (MF) and classify the images using Multi-SVM classifier. The entire system developed using python programming.

Bi-directional Block Encoding for Reversible Data Hiding over Encrypted Images

Reversible data hiding over encrypted images (RDH-EI) technology is a viable solution for privacy-preserving cloud storage, as it enables the reversible embedding of additional data into images while maintaining image confidentiality. Since the data hiders, e.g., cloud servers, are willing to embed as much data as possible for storage, management, or other processing purposes, a large embedding capacity is desirable in an RDH-EI scheme. In this article, we introduce a novel bi-directional block encoding (BDBE) method, which, for the first time, encodes the distances of values in a binary sequence from both ends. This approach allows for encoding images with smaller sizes compared to traditional and state-of-the-art encoding methods. Leveraging the BDBE technique, we propose a high-capacity RDH-EI scheme. In this scheme, the content owner initially predicts the image pixels and then employs BDBE to encode the prediction errors, creating space for data embedding. The resulting encoded data are subsequently encrypted using a secure stream cipher, such as the Advanced Encryption Standard, before being transmitted to a data hider. The data hider can embed confidential information within the encrypted image for the purposes of storage, management, or other processing. Upon receiving the data, an authorized receiver can accurately recover the original image and the embedded data without any loss. Experimental results demonstrate that our RDH-EI scheme achieves a significantly larger embedding capacity compared to several state-of-the-art schemes.

An unrestricted Arnold’s cat map transformation

AbstractThe Arnold’s Cat Map (ACM) is one of the chaotic transformations, which is utilized by numerous scrambling and encryption algorithms in Information Security. Traditionally, the ACM is used in image scrambling whereby repeated application of the ACM matrix, any image can be scrambled. The transformation obtained by the ACM matrix is periodic; therefore, the original image can be reconstructed using the scrambled image whenever the elements of the matrix, hence the key, is known. The transformation matrices in all the chaotic maps employing ACM has limitations on the choice of the free parameters which generally require the area-preserving property of the matrix used in transformation, that is, the determinant of the transformation matrix to be $$\pm 1.$$ ± 1 . This reduces the number of possible set of keys which leads to discovering the ACM matrix in encryption algorithms using the brute-force method. Additionally, the period obtained is small which also causes the faster discovery of the original image by repeated application of the matrix. These two parameters are important in a brute-force attack to find out the original image from a scrambled one. The objective of the present study is to increase the key space of the ACM matrix, hence increase the security of the scrambling process and make a brute-force attack more difficult. It is proved mathematically that area-preserving property of the traditional matrix is not required for the matrix to be used in scrambling process. Removing the restriction enlarges the maximum possible key space and, in many cases, increases the period as well. Additionally, it is supplied experimentally that, in scrambling images, the new ACM matrix is equivalent or better compared to the traditional one with longer periods. Consequently, the encryption techniques with ACM become more robust compared to the traditional ones. The new ACM matrix is compatible with all algorithms that utilized the original matrix. In this novel contribution, we proved that the traditional enforcement of the determinant of the ACM matrix to be one is redundant and can be removed.

A Regional Brightness Control Method for a Beam Projector to Avoid Human Glare

In this study, we proposed a system to reduce the speaker’s suffering from the strong light of a beam projector by applying regional brightness control over the screen. Since the original image and the projected one on the screen are quite different in area, brightness, and color, the proposed system first transforms them so that they have the same area and similar color tone. Then, to accurately determine the difference between those images, we have introduced a SSIM map, which is a perception-based method of measuring image similarity. Accordingly, an image segmentation model is used to determine the speaker’s silhouette from the SSIM map. We applied a couple of well-trained segmentation models, such as Selfie and DeepLab-v3, provided with MediaPipe. The experimental results showed the operability of the proposed system and that it determines most of a lecturer’s body area on the screen. To closely evaluate the system’s effectiveness, we have measured error rates consisting of false-positive and false-negative errors in the confusion matrix. With the measured results, the error rates appeared so insignificant and stable that the proposed system provides a practical effect for the speakers, especially in the case of applying DeepLab-v3. With the results, it is implied that an accurate segmentation model can considerably elevate the effectiveness of the system.

Improvement of lithium battery corner detection accuracy based on image restoration method

Target detection technology has been widely used in the automatic production of lithium batteries. However, motion blur will lead to the reduction of the angular position detection accuracy of lithium batteries. To solve this problem, an improved fuzzy recovery model for angular position of lithium battery is proposed in this paper. Firstly, the improved lightweight neural network RepVGG was used as the main module of the backbone network, so that the network could improve the performance of network feature extraction while reducing the number of calculation parameters and improving the reasoning speed of fuzzy restoration. Secondly, we optimize the multi-Dconv head transposed attention (MDTA) module and reference it to the generator, which reduces the complexity of the model and strengthens the network’s attention to details and textures, and improves the visual effect of the restored image. Finally, we design a lightweight globally connectable residual network called SAC Block and use it to to improve the discriminator, which enhances the global receptive field of the model and improves the structural similarity between the restored image and the original image. In order to verify the effectiveness of the method, we verify it on the self-built dataset and GoPro dataset. The experiments show that our proposed lightweight model improves the peak signal-to-noise ratio (PSNR) index by 9.2% and 8.6% respectively compared with the original model. The visual effect of the restored image is better than that of other current similar algorithms, and it is confirmed that our model can better improve the accuracy of lithium battery angular position detection.

Design and development of artificial intelligence‐based application programming interface for early detection and diagnosis of colorectal cancer from wireless capsule endoscopy images

AbstractThe colorectal cancer (CRC) is gaining attention in the context of gastrointestinal tract diseases as it ranks third among the most prevalent type of cancer. The early diagnosis of the CRC can be done by periodic examination of the colon and rectum for innocuous tissue abnormality called polyp as it has the potential to evolve as malignant in future. The CRC diagnosis using wireless capsule endoscopy requires the dedicated commitment of the medical expert demanding significant time, focus and effort. The accuracy of manual analysis in identifying polyps is extensively reliant on the cognitive condition of the physician, thus emphasizing the requirement for automatic polyp identification. The artificial intelligence integrated computer‐aided diagnosis system could assist the clinician in better diagnosis, thereby reducing the miss‐rates of polyps. In our proposed study, we developed an application program interface to aid the clinician in automatic segmentation of the polyp and evaluation of its dimension by manual placement of four landmarks on the predicted polyp. The segmentation is performed by the proposed light weight Padded U‐Net for the effective polyp segmentation in the colorectal images. We trained and validated the Padded U‐Net with augmented images of Kvasir dataset and calculated the performance parameters. In order to facilitate image augmentation, a graphical user interface called Augment Tree was developed, which incorporates 92 augmentation techniques. The accuracy, recall, precision, IoU, F1‐score, loss achieved during validation of Padded U‐Net were 95.6%, 0.946%, 0.985%, 0.933%, 0.965% and 0.080% respectively. We demonstrated that accuracy was improved and loss was reduced when the model was trained with augmented images rather than only the limited original dataset images. On comparison of our Padded U‐net architecture with recently developed architectures, our model attained optimal performance in all the metrics except accuracy in which it attained marginal performance to the highest value.

Deep-GD: Deep Learning based Automatic Garment Defect Detection and Type Classification

Garment defect detection has been successfully implemented in the quality quick response system for textile manufacturing automation. Defects in the production of textiles waste a lot of resources and reduce the quality of the finished goods. It is challenging to detect garment defects automatically because of the complexity of images and variety of patterns in textiles. This study presented a novel deep learning-based Garment defect detection framework named as Deep-GD model for sequentially identifying image defects in patterned garments and classify the defect types. Initially, the images are gathered from the HKBU database and bilateral filters are used in the pre-processing of images to remove distortions. A squeeze-and-excitation network (SE-net) combined with Random Decision Forest Classifier with Bayesian Optimization (RDF-BO) algorithm is used to detect and classify garment defects. By analyzing the differences among the original and reconstruction images, the unsupervised technique trains to rebuild the fabric pattern. The SE-net is used to identify certain fabric flaws in the garments of the pre-processed images. Then, the defects-related garments are processed using RDF-BO algorithm for classifying the garment defect from these regions. Finally, the proposed Deep-GD model is used for classifying defected fabric into 12 classes such as Defect free, soil stain, oil stain, Double end, Snarls, Miss, Horizontal stripes, Lumpy, Dye spot, Fall out, Hairiness, tiny hole. The proposed Deep-GD model achieves the overall classification accuracy of 97.16%, which is comparatively superior to the existing techniques.

Comparison and analysis of the number of convolutional layers in CNN-based breast cancer detection

Breast cancer is becoming more and more common, and the mortality rate is increasing, which is indeed a serious problem. To alleviate this problem, this study applies a convolutional neural network (CNN) model to analyze breast cancer images formed by mammography. The CNN model has the property of automatically learning hierarchical feature representations from the original image data. The structure of the CNN enables it to capture spatially hierarchical features at multiple scales, including edges, textures, and objects. Specifically, this study explores the performance of CNN models on prediction tasks by varying the number of convolutional layers. This study is conducted on the DDSM mammography dataset. The experimental results demonstrate that CNN models are effective for detecting breast cancer, both in terms of accuracy and loss rate, and that more convolutional layers improve performance. In particular, the model saturates at four convolutional layers to reach the highest performance. Thus, this study helps to accelerate the efficiency of breast cancer detection and paves the way for more efficient methods in the future.

An image encryption algorithm based on multi-layered chaotic maps and its security analysis

This study introduces a novel symmetric image encryption algorithm that employs a multi-layered architecture of chaotic maps, incorporating both confusion and diffusion components. The selected chaotic maps, namely Aizawa, Ricker, Sine-Circle, and Chirikov, have been well-established as effective sources of chaos in the existing literature. Within the proposed algorithm, these chaotic maps are strategically utilised in distinct layers to enhance the security of the encrypted image. The algorithm demonstrates applicability to real-world scenarios, distinguished by its simplicity, high-speed performance, and low computational complexity. The efficacy of the method is underlined by its dependence on both original images and secret keys, rendering it resilient against brute force and differential attacks. Rigorous experimentation, encompassing correlation, histogram analysis, NBCR, NPCR, UACI, PSNR, and NIST values, establishes the proposed encryption algorithm's robustness and security against diverse attack modalities, including differential and statistical approaches. These findings position the algorithm as a compelling solution for safeguarding sensitive information in various practical applications.

Ref-MEF: Reference-Guided Flexible Gated Image Reconstruction Network for Multi-Exposure Image Fusion.

Multi-exposure image fusion (MEF) is a computational approach that amalgamates multiple images, each captured at varying exposure levels, into a singular, high-quality image that faithfully encapsulates the visual information from all the contributing images. Deep learning-based MEF methodologies often confront obstacles due to the inherent inflexibilities of neural network structures, presenting difficulties in dynamically handling an unpredictable amount of exposure inputs. In response to this challenge, we introduce Ref-MEF, a method for color image multi-exposure fusion guided by a reference image designed to deal with an uncertain amount of inputs. We establish a reference-guided exposure correction (REC) module based on channel attention and spatial attention, which can correct input features and enhance pre-extraction features. The exposure-guided feature fusion (EGFF) module combines original image information and uses Gaussian filter weights for feature fusion while keeping the feature dimensions constant. The image reconstruction is completed through a gated context aggregation network (GCAN) and global residual learning GRL. Our refined loss function incorporates gradient fidelity, producing high dynamic range images that are rich in detail and demonstrate superior visual quality. In evaluation metrics focused on image features, our method exhibits significant superiority and leads in holistic assessments as well. It is worth emphasizing that as the number of input images increases, our algorithm exhibits notable computational efficiency.