Matching Image Pairs Research Articles

LightSGM: Local feature matching with lightweight seeded

Addressing the quintessential challenge of local feature matching in computer vision, this study introduces a novel fast sparse seed graph structure named LightSGM. This structure aims to refine the characterization of graph features while minimizing superfluous connections. Initially, a subset of high-quality seed feature points is curated using a confidence filter. Subsequently, keypoint features are assimilated into this seed set via graph pooling, and the composite features are further processed through a memory and computation-efficient seed transformer to capture rich contextual information about the keypoints. The seed feature points are then relayed back to the original keypoints using an inverse process known as graph unpooling. The paper also introduce an adaptive mechanism to determine the optimal number of model layers based on the intricacy of matching image pairs. A Matching Point Prediction Header is employed to extract the final set of matching points. Through extensive experimentation on image matching and position estimation, LightSGM has demonstrated its prowess in delivering competitive matching accuracy while maintaining a balance with real-time processing capabilities.

Unsupervised low-dose CT denoising using bidirectional contrastive network

Background and Objective:Low-dose computed tomography (LDCT) scans significantly reduce radiation exposure, but introduce higher levels of noise and artifacts that compromise image quality and diagnostic accuracy. Supervised learning methods have proven effective in denoising LDCT images, but are hampered by the need for large, paired datasets, which pose significant challenges in data acquisition. This study aims to develop a robust unsupervised LDCT denoising method that overcomes the reliance on paired LDCT and normal-dose CT (NDCT) samples, paving the way for more accessible and practical denoising techniques. Methods:We propose a novel unsupervised network model, Bidirectional Contrastive Unsupervised Denoising (BCUD), for LDCT denoising. This model innovatively combines a bidirectional network structure with contrastive learning theory to map the precise mutual correspondence between the noisy LDCT image domain and the clean NDCT image domain. Specifically, we employ dual encoders and discriminators for domain-specific data generation, and use unique projection heads for each domain to adaptively learn customized embedded representations. We then align corresponding features across domains within the learned embedding spaces to achieve effective noise reduction. This approach fundamentally improves the model’s ability to match features in latent space, thereby improving noise reduction while preserving fine image detail. Results:Through extensive experimental validation on the AAPM-Mayo public dataset and real-world clinical datasets, the proposed BCUD method demonstrated superior performance. It achieved a peak signal-to-noise ratio (PSNR) of 31.387 dB, a structural similarity index measure (SSIM) of 0.886, an information fidelity criterion (IFC) of 2.305, and a visual information fidelity (VIF) of 0.373. Notably, subjective evaluation by radiologists resulted in a mean score of 4.23, highlighting its advantages over existing methods in terms of clinical applicability. Conclusions:This paper presents an innovative unsupervised LDCT denoising method using a bidirectional contrastive network, which greatly improves clinical applicability by eliminating the need for perfectly matched image pairs. The method sets a new benchmark in unsupervised LDCT image denoising, excelling in noise reduction and preservation of fine structural details.

Semantic Distance Adversarial Learning for Text-to-Image Synthesis

Text-to-Image (T2I) synthesis is a cross-modality task that requires a text description as input to generate a realistic and semantically consistent image. To guarantee semantic consis- tency, previous studies regenerate text descriptions from synthetic images and align them with the given descriptions. However, the existing redescription modules lack explicit modeling of their training objectives, which is crucial for reliable measurement of semantic distance between redescriptions and given text inputs. Consequently, the aligned text redescriptions suffer from training bias caused by the emergence of adversarial image samples, unseen semantics, and mistaken contents from low- quality synthesized images. To this end, we propose a SEMantic distance Adversarial learning (SEMA) framework for Text-to- Image synthesis which strengthens semantic consistency from two aspects: 1) We introduce adversarial learning between the image generator and the text redescription module to mutually promote or demote the quality of generated image or text instances. This learning model ensures accurate redescription of image contents, thus diminishing the generation of adversarial image samples. 2) We introduce two-fold semantic distance discrimination (SEM distance) to characterize semantic relevance between matching text or image pairs. The unseen semantics and mistaken contents will be penalized with a large SEM distance. The proposed discrimination method also simplifies the model training process with no need to optimize multiple discriminators. Experimental results on CUB Birds 200 and MS-COCO datasets show that the proposed model outperforms the state-of-the-art methods. Code is available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/NJUPT-MCC/SEMA</uri> .

Node2Node: Self-Supervised Cardiac Diffusion Tensor Image Denoising Method

Although the existing cardiac diffusion tensor imaging (DTI) denoising methods have achieved promising results, most of them are dependent on the number of diffusion gradient directions, noise distributions, and noise levels. To address these issues, we propose a novel self-supervised cardiac DTI denoising network, Node2Node, which firstly expresses the diffusion-weighted (DW) image volumes along different directions as a graph, then the graph framelet transform (GFT) is implemented to map the DW signals into the GFT coefficients at different spectral bands, allowing us to accurately match the DW image pairs. After that, using the matched image pairs as input and target, a ResNet-like network is used to denoise in a self-supervised manner. In addition, a novel edge-aware loss based on pooling operation is proposed to retain the edge. Through comparison with several state-of-the-art methods on synthetic, ex vivo porcine, and in vivo human cardiac DTI datasets, we showed that the root mean square error (RMSE) of DW images and the average angular error (AAE) of fiber orientations obtained using Node2Node are the smallest, improved by 47.5% and 23.7%, respectively, on the synthetic dataset, demonstrating that Node2Node is not sensitive to the properties of the dataset.

Empirical mode decomposition and local binary pattern based feature extraction for face recognition

ABSTRACT Facial recognition is a challenging pattern recognition problem in computer vision. This paper proposes a face recognition system that uses Empirical Mode Decomposition (EMD) and Local Binary Pattern (LBP) based feature extraction for a robust face recognition system. This scheme initially decomposes the image into 2N number of IMF (Intrinsic Mode Function) images, where N numbers of IMF images are estimated in the X direction and N number of IMF images are estimated in the Y direction. From the 2N number of IMFs, the M number of best matching IMF image pairs is estimated using 2D Discrete Fourier Transform (DFT). The IMF pair is added to extract the IC-LBP (Intensity Compensated-Local Binary Pattern) features. The IC-LBP features are extracted from the IMF images such that the center intensity is adjusted based on an adaptive intensity threshold. The use of EMD and IC-LBP features uses the essential descriptors that best represent a facial image. The same process is repeated in the testing phase where the test image is categorized from the trained images using the Naïve Bayes algorithm. The performance evaluation was done using the Yale, MORPH, and FGNET using metrics such as time complexity and recognition rate on different types of test face images like partial faces, different lightning, and rotation. Results show that the proposed face recognition system outperforms the traditional algorithms. Results show that the proposed face recognition system outperforms the traditional algorithms.

Semi-supervised super-resolution of diffusion-weighted images based on multiple references.

Spatial resolution of diffusion tensor images is usually compromised to accelerate the acquisitions, and the state-of-the-art (SOTA) image super-resolution (SR) reconstruction methods are commonly based on supervised learning models. Considering that matched low-resolution (LR) and high-resolution (HR) diffusion-weighted (DW) image pairs are not readily available, we propose a semi-supervised DW image SR reconstruction method based on multiple references (MRSR) extracted from other subjects. In MRSR, the prior information of multiple HR reference images is migrated into a residual-like network to assist SR reconstruction of DW images, and a CycleGAN-based semi-supervised strategy is used to train the network with 30% matched and 70% unmatched LR-HR image pairs. We evaluate the performance of the MRSR by comparing against SOTA methods on an HCP dataset in terms of the quality of reconstructed DW images and diffusion metrics. MRSR achieves the best performance, with the mean PSNR/SSIM of DW images being improved by at least 14.3%/28.8% and 1%/1.4% respectively relative to SOTA unsupervised and supervised learning methods, and with the fiber orientations deviating from the ground truth by about 6.28° on average, the RMSEs of fractional anisotropy, mean diffusivity, axial diffusivity and radial diffusivity being 3.0%, 4.6%, 5.7% and 4.5% respectively relative to the ground truth. We validate the effectiveness of the proposed network structure, multiple-reference and CycleGAN-based semi-supervised learning strategies for SR reconstruction of diffusion tensor images through the ablation studies. The proposed method allows us to achieve SR reconstruction for diffusion tensor images with a limited number of matched image pairs.

Segmentation Agreement and AI-Based Feature Extraction of Cutaneous Infrared Images of the Obese Abdomen after Caesarean Section: Results from a Single Training Session

Background: Infrared thermography in women undergoing caesarean section has promise to identify a surgical site infection prodrome characterised by changes in cutaneous perfusion with concomitant influences on temperature distribution across the abdomen. This study was designed to compare abdominal and wound regions of interest (ROI) and feature extraction agreement between two independent users after a single training session. Methods: Image analysis performed manually in MATLAB with each reviewer ‘blind’ to results of the other. Image ROIs were annotated via pixel-level segmentation creating pixel masks at four time-points during the first 30 days after surgery. Results: A total of 366 matched image pairs (732 wound and abdomen labels in total) were obtained. Distribution of mask agreement using Jacquard similarity co-efficient ranged from 0.35 to 1. Good segmentation agreement (coefficient ≥ 0.7) (for mask size and shape) was observed for abdomen, but poor for wound (coefficient &lt; 0.7). From feature extraction, wound cold spots were observed most in those who later developed wound infections. Conclusions: Reviewer performance, with respect to the input (image) data in the first stage of algorithm development, reveals a lack of correspondence (agreement) of the ROI indicating the need for further work to refine the characteristics of output labels (masks) before an unsupervised algorithm works effectively to learn patterns and features of the wound.

OCT sequence registration before and after percutaneous coronary intervention (stent implantation)

To assess the progression of coronary artery disease, Optical Coherence Tomography (OCT) pullbacks acquired at different timepoints should be compared. However, the assessment of temporal sequences is a difficult task, as motion artifacts in the longitudinal and axial plane can decrease the quality of the manual inspection. To address this challenge, the current study presents a two-stage computational framework for the longitudinal and axial registration of two OCT pullbacks. During the first stage of the process, we focus on the accurate detection of the matching image pairs from the respective series, while during the second stage we focus on the axial registration of the matched pairs so that their common features are aligned. The dataset used includes 19 patients from two clinical centers, with two OCT pullbacks per patient: one before the stent implantation procedure and one after it. We applied our method on a synthetic dataset of OCT pullbacks, which was generated based on the in-vivo OCT pullbacks to reproduce the motion artifacts across the planes. In addition, the proposed method was validated on the 19 pairs of in-vivo OCT pullbacks with annotated pre/post stent deployment data. The method was able to reduce the alignment error from 32.17±26.14 to 5.6±6.6 frames, the rotational error from 11.59°±11.22° to 1.18°±0.81° and the distance error from 3.07mm±1.52mm to 0.46mm±0.44mm. In addition, the mean Mutual Information similarity increased by 13.47% after the longitudinal registration and an additional 123.33% after the axial registration on top of the previous one.

P15.08.B Intraoperative confocal laser endomicroscopy of brain tumors - potential and challenges from a neuropathological perspective

Abstract Background Intraoperative consultation, usually in the form of a frozen section, is an integral part of current neuropathology practice. Typical indications include determination of specimen adequacy, intraoperative diagnosis and distinction between tumoral and non-tumoral tissue in diffusely infiltrating gliomas in tumor periphery. The latter is currently limited by the fact that typically not all potential regions of interest can be sampled for frozen sections and the turnaround times of frozen sections usually do not permit multiple repetitive assessments. The commercial availability of a clinical-grade in vivo laser endomicroscope for neurosurgical applications may have the potential to change this and to permit a microscopic analysis in near-real time, which in turn may have a significant impact on resection strategies for brain tumors. In this work, we investigate the potential role of CLE in neurosurgery and the transferability of histological criteria to confocal imaging. Material and Methods The potential role of CLE in neurosurgery in comparison to other techniques for intraoperative tumor assessment was determined by analyzing common intraoperative neuropathological workflows and identifying unaddressed opportunities. The transferability of the histological criteria from standard histology to confocal imaging was achieved by analyzing matched image pairs (standard histology and CLE) from several common tumor entities. Results We identified the time gap between specimen resection and the availability of the frozen section results as an important application for intraoperative CLE imaging in neurosurgery. Importantly, neuropathologists could assess tissue in near real time in an unlimited number of digital biopsies prior to resection, which adds a potent new tool to the neurosurgical armamentarium. CLE images revealed features similar to already known tissue architecture/morphology seen in standard histology, which correlated well in matched histological images. Unique aspects of individual cells and other surrounding tumor tissue elements were observed in vivo while the CLE probe acquired images throughout its focal-depth range. Conclusion Widespread application of CLE may have a major impact on neuropathologists' role in intraoperative diagnosis and bring along both opportunities and challenges. For broad clinical adoption, strategies for histological criteria validation as well as for determination of diagnostic accuracy need to be developed.

Simulation-driven learning: a deep learning approach for image scanning microscopy via physical imaging models.

Image reconstruction based on deep learning has become an effective tool in fluorescence microscopy. Most deep learning reconstruction methods ignore the mechanism of the imaging process where a large number of datasets are required. In addition, a lot of time is spent solving the aliasing problem from multi-scaled image pairs for data pre-processing. Here we demonstrate an improved generative adversarial network for image scanning microscopy (ISM) that can be trained by simulation data and has good generalization. Based on physical imaging models, this method can generate matching image pairs from simulation images and uses them as datasets for network training, without capturing a large number of real ISM images and avoiding image alignment preprocessing. Simulation and experimental results show that this simulation data-driven method improves the imaging quality of conventional microscopic images and reduces the cost of experiments. This method provides inspiration for optimizing network generalizability of the deep learning network.

3MRS: An Effective Coarse-to-Fine Matching Method for Multimodal Remote Sensing Imagery

The fusion of image data from multiple sensors is crucial for many applications. However, there are significant nonlinear intensity deformations between images from different kinds of sensors, leading to matching failure. To address this need, this paper proposes an effective coarse-to-fine matching method for multimodal remote sensing images (3MRS). In the coarse matching stage, feature points are first detected on a maximum moment map calculated with a phase congruency model. Then, feature description is conducted using an index map constructed by finding the index of the maximum value in all orientations of convolved images obtained using a set of log-Gabor filters. At last, several matches are built through image matching and outlier removal, which can be used to estimate a reliable affine transformation model between the images. In the stage of fine matching, we develop a novel template matching method based on the log-Gabor convolution image sequence and match the template features with a 3D phase correlation matching strategy, given that the initial correspondences are achieved with the estimated transformation. Results show that compared with SIFT, and three state-of-the-art methods designed for multimodal image matching, PSO-SIFT, HAPCG, and RIFT, only 3MRS successfully matched all six types of multimodal remote sensing image pairs: optical–optical, optical–infrared, optical–depth, optical–map, optical–SAR, and day–night, with each including ten different image pairs. On average, the number of correct matches (NCM) of 3MRS was 164.47, 123.91, 4.88, and 4.33 times that of SIFT, PSO-SIFT, HAPCG, and RIFT for the successfully matched image pairs of each method. In terms of accuracy, the root-mean-square error of correct matches for 3MRS, SIFT, PSO-SIFT, HAPCG, and RIFT are 1.47, 1.98, 1.79, 2.83, and 2.45 pixels, respectively, revealing that 3MRS got the highest accuracy. Even though the total running time of 3MRS was the longest, the efficiency for obtaining one correct match is the highest considering the most significant number of matches. The source code of 3MRS and the experimental datasets and detailed results are publicly available.

Feature Matching Combining Radiometric and Geometric Characteristics of Images, Applied to Oblique- and Nadir-Looking Visible and TIR Sensors of UAV Imagery.

A large amount of information needs to be identified and produced during the process of promoting projects of interest. Thermal infrared (TIR) images are extensively used because they can provide information that cannot be extracted from visible images. In particular, TIR oblique images facilitate the acquisition of information of a building’s facade that is challenging to obtain from a nadir image. When a TIR oblique image and the 3D information acquired from conventional visible nadir imagery are combined, a great synergy for identifying surface information can be created. However, it is an onerous task to match common points in the images. In this study, a robust matching method of image pairs combined with different wavelengths and geometries (i.e., visible nadir-looking vs. TIR oblique, and visible oblique vs. TIR nadir-looking) is proposed. Three main processes of phase congruency, histogram matching, and Image Matching by Affine Simulation (IMAS) were adjusted to accommodate the radiometric and geometric differences of matched image pairs. The method was applied to Unmanned Aerial Vehicle (UAV) images of building and non-building areas. The results were compared with frequently used matching techniques, such as scale-invariant feature transform (SIFT), speeded-up robust features (SURF), synthetic aperture radar–SIFT (SAR–SIFT), and Affine SIFT (ASIFT). The method outperforms other matching methods in root mean square error (RMSE) and matching performance (matched and not matched). The proposed method is believed to be a reliable solution for pinpointing surface information through image matching with different geometries obtained via TIR and visible sensors.

Selecting image pairs for structure-from-motion by introducing the image spatial position

ABSTRACT Image matching is a quite time-consuming task for Structure-from-Motion (SfM). In this paper, a Bag-of-Words (BoW) model that reduces the feature dimensions and introduces image spatial locations is proposed to improve the efficiency and reliability of SfM. The whole workflow includes three steps. Firstly, principal component analysis (PCA) is used to reduce the high-dimensional features to low-dimensional features, so as to improve the efficiency of retrieval vocabulary construction. Secondly, by calculating the inverse distance weighting score of query images, a comprehensive retrieval score is constructed to improve the distinguishability between similar images. Finally, by calculating the retrieval threshold and discarding the invalid matching image pairs, the image query precision is further improved. The experimental results show that compared with the VocabTree (VT) and the Hamming Embedding (HE) methods, the proposed algorithm for image matching time is reduced by 69.5% and 72.0%, respectively, while the number of sparse point clouds of the reconstruction is increased 0.6%.

NCNet: Neighbourhood Consensus Networks for Estimating Image Correspondences

We address the problem of finding reliable dense correspondences between a pair of images. This is a challenging task due to strong appearance differences between the corresponding scene elements and ambiguities generated by repetitive patterns. The contributions of this work are threefold. First, inspired by the classic idea of disambiguating feature matches using semi-local constraints, we develop an end-to-end trainable convolutional neural network architecture that identifies sets of spatially consistent matches by analyzing neighbourhood consensus patterns in the 4D space of all possible correspondences between a pair of images without the need for a global geometric model. Second, we demonstrate that the model can be trained effectively from weak supervision in the form of matching and non-matching image pairs without the need for costly manual annotation of point to point correspondences. Third, we show the proposed neighbourhood consensus network can be applied to a range of matching tasks including both category- and instance-level matching, obtaining the state-of-the-art results on the PF, TSS, InLoc, and HPatches benchmarks.

Incoherent Clutter Suppression Using Lag-One Coherence.

The lag-one coherence (LOC), derived from the correlation between the nearest-neighbor channel signals, provides a reliable measure of clutter which, under certain assumptions, can be directly related to the signal-to-noise ratio of individual channel signals. This offers a direct means to decompose the beamsum output power into contributions from speckle and spatially incoherent noise originating from acoustic clutter and thermal noise. In this study, we applied a novel method called lag-one spatial coherence adaptive normalization (LoSCAN) to locally estimate and compensate for the contribution of spatially incoherent clutter from conventional delay-and-sum (DAS) images. Suppression of incoherent clutter by LoSCAN resulted in improved image quality without introducing many of the artifacts common to other adaptive imaging methods. In simulations with known targets and added channel noise, LoSCAN was shown to restore native contrast and increase DAS dynamic range by as much as 10-15 dB. These improvements were accompanied by DAS-like speckle texture along with reduced focal dependence and artifact compared with other adaptive methods. Under in vivo liver and fetal imaging conditions, LoSCAN resulted in increased generalized contrast-to-noise ratio (gCNR) in nearly all matched image pairs ( N = 366 ) with average increases of 0.01, 0.03, and 0.05 in good-, fair-, and poor-quality DAS images, respectively, and overall changes in gCNR from -0.01 to 0.20, contrast-to-noise ratio (CNR) from -0.05 to 0.34, contrast from -9.5 to -0.1 dB, and texture μ/σ from -0.37 to -0.001 relative to DAS.

Anomaly Detection of Trackside Equipment Based on GPS and Image Matching

The health of trackside equipment often seriously affects the safe driving of the train, while manually checking their status is an indispensably laborious task. In order to automate the task, this paper addresses the problem of automatically and accurately detecting anomalies of equipment located along the track through GPS and image calibration techniques. Considering the unnoticeable changes in equipment, including screws missing and cable corrosion, detection by classification-based machine learning methods is difficult to implement, thus in this paper we propose and conduct a novel detection mechanism by an efficient way of image subtraction. Especially, our method consists of four steps. The first step is to collect images of the trackside equipment and their corresponding mileage in the same route several times through cameras and GPS devices installed in the inspection train. Then, by introducing an improved RANSAC algorithm, the GPS data is further corrected. Secondly, we define one pass of the collected image data along with GPS information as template, and for the rest passes that need to be detected, the first frame alignment operation is operated through GPS and image information. For the third step, the SURF algorithm is implemented for image matching, and then the subtraction operation is conducted between each matched image pairs. Finally, we use empirical filtering mechanism to remove false positives that we have detected. Experimental results demonstrate the efficiency and effectiveness of our proposed method for anomalies detection of trackside equipment.

An Unordered Image Stitching Method Based on Binary Tree and Estimated Overlapping Area.

Aiming at the complex computation and time-consuming problem during unordered image stitching, we present a method based on the binary tree and the estimated overlapping areas to stitch images without order in this paper. For image registration, the overlapping areas between input images are estimated, so that the extraction and matching of feature points are only performed in these areas. For image stitching, we build a model of the binary tree to stitch each two matched images without sorting. Compared to traditional methods, our method significantly reduces the computational time of matching irrelevant image pairs and improves the efficiency of image registration and stitching. Moreover, the stitching model of the binary tree proposed in this paper further reduces the distortion of the panorama. Experimental results show that the number of extracted feature points in the estimated overlapping area is approximately 0.3∼0.6 times of that in the entire image by using the same method, which greatly reduces the computational time of feature extraction and matching. Compared to the exhaustive image matching method, our approach only takes about 1/3 of the time to find all matching images.

Near Duplicate Image Retrieval using Multilevel Local and Global Convolutional Neural Network Features

In this work, we present an approach based on multilevel local as well as global Convolutional Neural Network (CNN) feature matching to retrieve near duplicate images. CNN features are suitable for visual matching. The CNN features of entire image may not give accuracy in retrieval due to various image editing/capturing operations. Our retrieval task focuses on matching image pairs based on local and global levels. In local matching, an image is segmented into fixed size blocks followed by extracting patches by considering neighboring regions at different levels. Matching local image patches at different levels provides robustness to our retrieval model. In local patch extraction, we select blocks containing SURF feature points instead of selecting all blocks. CNN features are extracted and stored for each image patch and then followed by extraction of global CNN features. Finally, similarity between image pairs is computed by considering all extracted CNN features. Our similarity function is based on correlation and number of blocks found in matching. We implemented our proposed approach on benchmarking Holiday dataset. Retrieval results show remarkable improvement in mean average precision (mAP) on the dataset.

A method for feature matching in images using descriptor structures

A method of feature matching in images using descriptor structures is considered in the work. The descriptors in the developed method can be any known solutions in the field of computer vision. However, inaccuracies can occur when matching image pairs. It is proposed that descriptor structures should be compared to eliminate the “outliers”. Descriptor structures are described using graphs. An Umeyama method is used to find matching features using descriptor structures. The method is based on the decomposition of matrices into eigenvalues and eigenvectors for weighted graph matching problems. Thus, matches are based on the descriptor at the initial stage. Two graphs are then constructed for each image based on the resulting sets of mapped features. The weights of the graph are distances between all image features, calculated using the Gauss function. Weight matrices are built for each graph. Matrix decomposition is carried out into eigenvalues and eigenvectors. The resulting matrix is calculated based on the Umeyama method and correct matches are found. Thus, false matches are excluded from the set of matches obtained using descriptors by comparing structures. The method is invariant to zoom and in-plane image rotation. The method leads to correct results only if the number of correct matches is greater than the number of false matches. The complexity of the developed algorithm is proportional to the number of matches found.

Contextual modeling on auxiliary points for robust image reranking

Image reranking is an effective post-processing step to adjust the similarity order in image retrieval. As key components of initialized ranking lists, top-ranked neighborhoods of a given query usually play important roles in constructing dissimilarity measure. However, the number of pertinent candidates varies with respect to different queries. Thus the images with short lists of ground truth suffer from insufficient contextual information. It consequently introduces noises when using k-nearest neighbor rule to define the context. In order to alleviate this problem, this paper proposes auxiliary points which are added as assistant neighbors in an unsupervised manner. These extra points act on revealing implicit similarity in the metric space and clustering matched image pairs. By isometrically embedding each constructed metric space into the Euclidean space, the image relationships on underlying topological manifolds are locally represented by distance descriptions. Furthermore, by combining Jaccard index with our auxiliary points, we present a contextual modeling on auxiliary points (CMAP) method for image reranking.With richer contextual activations, the Jaccard similarity coefficient defined by local distribution achieves more reliable outputs as well as more stable parameters. Extensive experiments demonstrate the robustness and effectiveness of the proposed method.