Visual Similarity Research Articles

The item-specific proportion congruency effect transfers to non-category members based on broad visual similarity.

The item-specific proportion congruency (ISPC) effect-that Stroop effects are reduced for items that are more likely to be incongruent than congruent-indicates that humans have the remarkable capacity to resolve conflict when it is associated with statistical regularities in the environment. It has been demonstrated that an ISPC signal induced by mostly congruent and mostly incongruent inducer items transfers to a set of distinct but visually similar transfer items that are equally likely to be congruent and incongruent; however, it is unclear what the ISPC signal is associated with to allow its transfer. To investigate this issue, an animal Stroop task was used to evaluate whether the ISPC signal would transfer to animal pictures that were different but visually similar same-category members (e.g., retrievers to retrievers, Experiment 1), visually dissimilar same-category members with broadly similar features (e.g., retrievers to bulldogs, Experiment 2), and visually dissimilar different-category members with broadly similar features (e.g., retrievers to house cats, Experiment 3). It was revealed that an ISPC effect was observed for the transfer items of each experiment, suggesting that these conflict signals can be linked based on broad feature similarity.

Searching for Signs: Developing a Handshape Taxonomy Based on Visual Similarity

Abstract When designing a sign language dictionary or lexical database, a criterion that needs to be decided on is how to organize handshapes in the search-by-sign interface. Although notation systems do exist for indexing and searching sign entries, dictionaries do not often use them. This study investigates how to classify and order images of handshapes without relying on the alphabetical order of handshape names or phonological parameters. For this purpose, a cluster analysis using three variables was applied, resulting in groups based on visual similarities between handshapes. The objects in the resulting clusters were reordered based on similarity and shape gradation principles, aiming for an optimal organization with regard to Prägnanz. Ultimately, as we expect native signers to rely on the visual features of signs, the taxonomy proposed in this study makes it possible to display handshapes in a way that requires fewer metalinguistic skills to support signers’ search actions.

Distributed densely connected convolutional network approach on patient’s metadata of dermoscopic images for early melanoma detection

Melanoma is one of the most deadly cancers on the planet, and it has the ability to spread to new places of the body if not discovered early enough. Dermoscopic pictures are routinely used to identify melanoma. Many previous studies, based on standard classification approaches and deep learning models, have been proposed for automated analysis of skin lesions. In traditional classification systems, handcrafted functions are used as input. However, due to the high visual similarity between different classes of skin lesions and complex skin diseases, hand-made features are not discriminating enough and fail in many circumstances. Convolutional networks with fewer connections between the input layers and those close to the output can be significantly deeper, more accurate and more effective for training, according to a recent study. In this study, we accept and offer the application of Hadoop Distributed Tight Convolution Network (HdiDenseNet) for melanoma skin cancer detection and study of failed lesions, which can be connected to the mode of transmission. You can use all feature maps at all input levels as input, and your own maps will be used as inputs in all poster headers.

Objective discrimination of fur type based on electrophoresis optimized for fur-hair proteins.

The accurate identification of animal species used for fur is important for conserving endangered animals, stopping illegal fur distribution, and addressing consumer concerns. Animal species used for fur are currently differentiated by observing species-specific morphological fur-hair features through a microscope. Although this method is simple, the results may differ among inspectors owing to its subjective nature. To develop an objective approach for differentiating animal species based on fur, we utilized the electrophoretic patterns of fur-hair proteins. First, we optimized protein extraction methods to produce clear electrophoretic patterns from fur-hair proteins. Then, we obtained 324 electrophoretic patterns from 54 fur samples belonging to 24 different animals; 216 of the 324 patterns were used for the construction of a discrimination model using two-way orthogonal partial least squares discriminant analysis. The model correctly discriminated between all the remaining 108 patterns without any false negatives or positives. Moreover, this model could discriminate between fur samples from closely related species that are difficult to distinguish using conventional microscopic identification because of the visual similarity of the fur hairs.

Data-driven de-smearing of DSC signals

In heat flux differential scanning calorimetry, a pair of identical crucibles, one empty as reference and the other filled with the sample, is heated in a furnace with a prescribed rate. The (empty) reference crucible heats faster, resulting in a temperature difference that is detected by thermocouples. Slow heating of the furnace results in weak and noisy signals; higher heating rates induce strong signals but lead to smearing if applied to materials undergoing a phase transition: The recorded peak signal is shifted toward higher temperatures. To determine the peak, the onset/endset temperatures, and the phase transition enthalpy, multiple heating rates are used to find a trade-off between noise and smearing. When plotting the melting peaks over temperature and heating rate, the visual similarity to the time evolution of a probability density under drift and diffusion catches the eye. Such a density evolution can be described by the Fokker–Planck equations. In this analogon, the de-smeared signal corresponds to the initial distribution. We propose a data-driven de-smearing approach, based on an extrapolation to a (hypothetical) zero heating rate signal. This zero rate signal is low-dimensionally parameterized and its parameters together with the drift and diffusion of the Fokker–Planck equation are fitted against the calorimetric measurements. The method is successfully tested on heat capacity data of a technical-grade high-density polyethylene (HDPE) using mid-range heating rates. The data are strongly affected by smearing, and the proposed de-smearing method {FPEX}_{0} delivers reliable estimates of characteristic shape parameters of the phase transition peak effectively overcoming the problem of a deteriorating signal-to-noise ratio for heating rates approaching zero.

A Deep Feature Aggregation Network for Accurate Indoor Camera Localization

As scene coordinate regression (SCoRe) methods become prevailing in the area of visual camera localization, the issue of repetitive or sparse texture scenes continues to be a concern. Specifically, they will suffer from performance degeneration due to ambiguous patterns caused by visual similarity. In this work, we propose a novel network for camera localization through a single RGB image, with our key insight that taking only high-level feature maps as input can be difficult for the network to accurately model the regression problem due to ambiguous patterns and utilizing the rich spatial details in low-level feature maps can tackle this issue. The core components of the network are 1) Deep Feature Aggregation Module (DFAM), which eliminates the difference among the different level feature representations and fuses multi-level context information; 2) CoordConv Scheme, which further improves the discrimination of features in repetitive or sparse texture areas of the image; 3) Deep Supervision, which endows low-level feature maps with direct supervision from the ground truth to improve the accuracy of camera localization; 4) Uncertainty Modeling, which quantifies the prediction errors stemming from the intrinsic noise in the data. Moreover, to maximize the power of DFAM, we embed channel attention modules into it to prune redundant and noisy features, through which we can refine the different level feature maps. Our network is designed to be lightweight and efficient, and the proposed DFAM can be integrated into general SCoRe-based networks. Comprehensive experiments demonstrate the effectiveness of DFAM and the superiority of our network over the state-of-the-art methods on two benchmarks.

Discriminative Cervical Lesion Detection in Colposcopic Images With Global Class Activation and Local Bin Excitation.

Accurate cervical lesion detection (CLD) methods using colposcopic images are highly demanded in computer-aided diagnosis (CAD) for automatic diagnosis of High-grade Squamous Intraepithelial Lesions (HSIL). However, compared to natural scene images, the specific characteristics of colposcopic images, such as low contrast, visual similarity, and ambiguous lesion boundaries, pose difficulties to accurately locating HSIL regions and also significantly impede the performance improvement of existing CLD approaches. To tackle these difficulties and better capture cervical lesions, we develop novel feature enhancing mechanisms from both global and local perspectives, and propose a new discriminative CLD framework, called CervixNet, with a Global Class Activation (GCA) module and a Local Bin Excitation (LBE) module. Specifically, the GCA module learns discriminative features by introducing an auxiliary classifier, and guides our model to focus on HSIL regions while ignoring noisy regions. It globally facilitates the feature extraction process and helps boost feature discriminability. Further, our LBE module excites lesion features in a local manner, and allows the lesion regions to be more fine-grained enhanced by explicitly modelling the inter-dependencies among bins of proposal feature. Extensive experiments on a number of 9888 clinical colposcopic images verify the superiority of our method (AP .75 = 20.45) over state-of-the-art models on four widely used metrics.

Classification of melanoma images using 2D multifractal detrended cross-correlation analysis

Skin cancer is a common human malignant tumor and melanoma is one of the most fatal diseases in skin cancer. There exists a high degree of visual similarity between melanoma and non-melanoma. In addition, the acquisition and labeling of skin cancer images need relevant medical knowledge, therefore, it is difficult to obtain natural images. These problems make it difficult to distinguish between melanoma and non-melanoma. How to extract the high-dimensional features of skin cancer images is the main problem to improve the classification performance of skin cancer images. For this purpose, we propose a new computer-aided classification system with a two-dimensional (2D) multifractal detrended fluctuation analysis (MF-DCCA) method and classifier combination. The proposed 2D MF-DCCA is expanded by the 2D MF-DFA and we aim to change the distribution of the generalized Hurst exponents calculated by the original 2D MF-DFA, making the image features represented by the generalized Hurst exponents calculated by 2D MF-DCCA are more significant. Therefore, we take the generalized Hurst exponents calculated by the two methods into the classifier and evaluate their classification performance. The classification metrics, such as Accuracy (Acc), Sensitivity (Sen), and Specificity (Spe) of two classifiers, show that the features extracted by MF-DCCA are better than those by MF-DFA. In addition, among the two classifiers such as the SVM and k-NN, the k-NN performs best in Acc and Spe with [Formula: see text] and [Formula: see text], respectively. The k-NN performs better than SVM in the classification Sen with [Formula: see text].

ARACAM: A RGB-D Multi-View Photogrammetry System for Lower Limb 3D Reconstruction Applications.

In the world, there is a growing need for lower limb prostheses due to a rising number of amputations caused primarily, by diabetic foot. Researchers enable functional and comfortable prostheses through prosthetic design by integrating new technologies applied to the traditional handcrafted method for prosthesis fabrication that is still current. That is why computer vision shows to be a promising tool for the integration of 3D reconstruction that may be useful for prosthetic design. This work has the objective to design, prototype, and test a functional system to scan plaster cast molds, which may serve as a platform for future technologies for lower limb reconstruction applications. The image capture system comprises 5 stereoscopic color and depth cameras, each with 4 DOF mountings on an enveloping frame, as well as algorithms for calibration, segmentation, registration, and surface reconstruction. The segmentation metrics of dice coefficient and Hausdorff distance (HD) show strong visual similarity with an average similarity of 87% and average error of 6.40 mm, respectively. Moving forward, the system was tested on a known 3D printed model obtained from a computer tomography scan to which comparison results via HD show an average error of ≤1.93 mm thereby making the system competitive against the systems reviewed from the state-of-the-art.

Activity in perirhinal and entorhinal cortex predicts perceived visual similarities among category exemplars with highest precision.

Vision neuroscience has made great strides in understanding the hierarchical organization of object representations along the ventral visual stream (VVS). How VVS representations capture fine-grained visual similarities between objects that observers subjectively perceive has received limited examination so far. In the current study, we addressed this question by focussing on perceived visual similarities among subordinate exemplars of real-world categories. We hypothesized that these perceived similarities are reflected with highest fidelity in neural activity patterns downstream from inferotemporal regions, namely in perirhinal (PrC) and anterolateral entorhinal cortex (alErC) in the medial temporal lobe. To address this issue with functional magnetic resonance imaging (fMRI), we administered a modified 1-back task that required discrimination between category exemplars as well as categorization. Further, we obtained observer-specific ratings of perceived visual similarities, which predicted behavioural discrimination performance during scanning. As anticipated, we found that activity patterns in PrC and alErC predicted the structure of perceived visual similarity relationships among category exemplars, including its observer-specific component, with higher precision than any other VVS region. Our findings provide new evidence that subjective aspects of object perception that rely on fine-grained visual differentiation are reflected with highest fidelity in the medial temporal lobe.

Modeling Hybrid Feature-Based Phishing Websites Detection Using Machine Learning Techniques

In this paper, we mainly present a machine learning based approach to detect real-time phishing websites by taking into account URL and hyperlink based hybrid features to achieve high accuracy without relying on any third-party systems. In phishing, the attackers typically try to deceive internet users by masking a webpage as an official genuine webpage to steal sensitive information such as usernames, passwords, social security numbers, credit card information, etc. Anti-phishing solutions like blacklist or whitelist, heuristic, and visual similarity based methods cannot detect zero-hour phishing attacks or brand-new websites. Moreover, earlier approaches are complex and unsuitable for real-time environments due to the dependency on third-party sources, such as a search engine. Hence, detecting recently developed phishing websites in a real-time environment is a great challenge in the domain of cybersecurity. To overcome these problems, this paper proposes a hybrid feature based anti-phishing strategy that extracts features from URL and hyperlink information of client-side only. We also develop a new dataset for the purpose of conducting experiments using popular machine learning classification techniques. Our experimental result shows that the proposed phishing detection approach is more effective having higher detection accuracy of 99.17% with the XG Boost technique than traditional approaches.

Neural Style Transfer for image within images and conditional GANs for destylization

In this paper, the feature representation of an image by CNN is used to hide the secret image into the cover image. The style of the cover image hides the content of the secret image and produce a stego image using Neural Style Transfer (NST) algorithm, which resembles the cover image and also contains the semantic content of secret image. The main technical contributions are to hide the content of the secret image in the in-between hidden layered style features of the cover image, which is the first of its kind in the present state-of-art-technique. Also, to recover the secret image from the stego image, destylization is done with the help of conditional generative adversarial networks (GANs) using Residual in Residual Dense Blocks (RRDBs). Further, stego images from different layer combinations of content and style features are obtained and evaluated. Evaluation is based on the visual similarity and quality loss between the cover-stego pair and the secret-reconstructed secret pair of images. From the experiments, it has been observed that the proposed algorithm has 43.95 dB Peak Signal-to-Noise Ratio (PSNR)), .995 Structural Similarity Index (SSIM), and .993 Visual Information Fidelity (VIF) for the ImageNet dataset. The proposed algorithm is found to be more robust against StegExpose than the traditional methods.

Detection of cervical lesions in colposcopic images based on the RetinaNet method

There is a critical requirement for detecting cervical lesions through colposcopic images in computer-aided diagnosis. Compared to images from natural scenes, colposcopic images have some specific problems, such as low contrast, high visual similarity, and blurry lesion boundaries that make it difficult to accurately detect cervical lesion areas. To solve these problems, this paper proposes a method based on RetinaNet to detect lesion areas in colposcopic images. First, the depth features of the entire image are extracted by a fusion of ResNet50 and a feature pyramid network (FPN). In addition, the model suppresses the weight of simple and easy-to-distinguish samples through local loss, ensuring that the training can focus on the hard-to-distinguish and that important samples, while improving the utilization rate of the important features. Then, object classification and bounding box regression are performed on the feature map through two subnets. Under the same experimental conditions, the detection effects of this method are compared with those of other mainstream models through the mean average precision (mAP), average recall (AR) and other indexes. Experimental results show that the method based on RetinaNet is superior to these compared models, with a mAP[.5:.95] of 32.72%, a mAP.5 of 50.16%, and an AR of 49.70%. Compared with those of Faster R-CNN-ResNet50 + FPN, the mAP[.5:.95] is 2.76% higher and the AR is 6.42% higher.

Diagnosis of Esophageal Lesions by Multi-Classification and Segmentation Using an Improved Multi-Task Deep Learning Model.

It is challenging for endoscopists to accurately detect esophageal lesions during gastrointestinal endoscopic screening due to visual similarities among different lesions in terms of shape, size, and texture among patients. Additionally, endoscopists are busy fighting esophageal lesions every day, hence the need to develop a computer-aided diagnostic tool to classify and segment the lesions at endoscopic images to reduce their burden. Therefore, we propose a multi-task classification and segmentation (MTCS) model, including the Esophageal Lesions Classification Network (ELCNet) and Esophageal Lesions Segmentation Network (ELSNet). The ELCNet was used to classify types of esophageal lesions, and the ELSNet was used to identify lesion regions. We created a dataset by collecting 805 esophageal images from 255 patients and 198 images from 64 patients to train and evaluate the MTCS model. Compared with other methods, the proposed not only achieved a high accuracy (93.43%) in classification but achieved a dice similarity coefficient (77.84%) in segmentation. In conclusion, the MTCS model can boost the performance of endoscopists in the detection of esophageal lesions as it can accurately multi-classify and segment the lesions and is a potential assistant for endoscopists to reduce the risk of oversight.

Are brand names special words? Letter visual-similarity affects the identification of brand names, but not common words.

Brand names are often considered a special type of words of special relevance to examine the role of visual codes during reading: unlike common words, brand names are typically presented with the same letter‐case configuration (e.g., IKEA, adidas). Recently, Pathak et al. (European Journal of Marketing, 2019, 53, 2109) found an effect of visual similarity for misspelled brand names when the participants had to decide whether the brand name was spelled correctly or not (e.g., tacebook [baseword: facebook] was responded more slowly and less accurately than xacebook). This finding is at odds with both orthographically based visual‐word recognition models and prior experiments using misspelled common words (e.g., viotin [baseword: violin] is identified as fast as viocin). To solve this puzzle, we designed two experiments in which the participants had to decide whether the presented item was written correctly. In Experiment 1, following a procedure similar to Pathak et al. (European Journal of Marketing, 2019, 53, 2109), we examined the effect of visual similarity on misspelled brand names with/without graphical information (e.g., anazon vs. atazon [baseword: amazon]). Experiment 2 was parallel to Experiment 1, but we focused on misspelled common words (e.g., anarillo vs. atarillo; baseword: amarillo [yellow in Spanish]). Results showed a sizeable effect of visual similarity on misspelled brand names – regardless of their graphical information, but not on misspelled common words. These findings suggest that visual codes play a greater role when identifying brand names than common words. We examined how models of visual‐word recognition can account for this dissociation.

Eliciting false insights with semantic priming

The insight experience (or ‘Aha moment’) generally evokes strong feelings of certainty and confidence. An ‘Aha’ experience for a false idea could underlie many false beliefs and delusions. However, for as long as insight experiences have been studied, false insights have remained difficult to elicit experimentally. That difficulty, in turn, highlights the fact that we know little about what causes people to experience a false insight. Across two experiments (total N = 300), we developed and tested a new paradigm to elicit false insights. In Experiment 1 we used a combination of semantic priming and visual similarity to elicit feelings of insight for incorrect solutions to anagrams. These false insights were relatively common but were experienced as weaker than correct ones. In Experiment 2 we replicated the findings of Experiment 1 and found that semantic priming and visual similarity interacted to produce false insights. These studies highlight the importance of misleading semantic processing and the feasibility of the solution in the generation of false insights.

Stylistic Traces of Amarna Art in Reliefs of the Tomb of Petosiris (Tuna al Gebel Necropolis)

Abstract The famous tomb of Petosiris (a high priest of Thoth) in the Tuna al Gebel necropolis on the outskirts of the Middle Egyptian town of Hermopolis Magna is dated by the last quarter of the 4th c. bc. The pictorial program of the tomb, decorated with painted reliefs, has no known parallels in Ptolemaic art. In terms of iconography and style the reliefs rely on both Egyptian and Greek artistic traditions. This study attempts to trace elements of the so-called Amarna style among the whole variety of cultural impacts that shaped the unique character of the reliefs. Not only the visual similarity but also the short distance between the necropoli in Hermopolis Magna and Amarna city support the impetus for looking for pictorial parallels between the reliefs of the Petosiris tomb and the Amarna monuments. Some distinct stylistic details of the reliefs find no matches in Egyptian art but for the Amarna period. Nor do they originate from Macedonian or provincial Greek art. The reliefs of the tomb of Petosiris seem to be the only extant piece of art of the Graeco-Roman period that goes back to Amarna tradition in terms of techniques, style and, to some extent, iconography. The styling of the reliefs relies on three distinct visual paradigms – Classical (namely archaic) Egyptian, Amarna and Hellenistic – and seems to depend on subject matters of the particular parts of the program. The stylistic features deriving from different traditions probably served as a special visual language that helped to convey important meanings and connotations.

Visual and semantic similarity norms for a photographic image stimulus set containing recognizable objects, animals and scenes.

We collected visual and semantic similarity norms for a set of photographic images comprising 120 recognizable objects/animals and 120 indoor/outdoor scenes. Human observers rated the similarity of pairs of images within four categories of stimuli-inanimate objects, animals, indoor scenes and outdoor scenes-via Amazon's Mechanical Turk. We performed multidimensional scaling (MDS) on the collected similarity ratings to visualize the perceived similarity for each image category, for both visual and semantic ratings. The MDS solutions revealed the expected similarity relationships between images within each category, along with intuitively sensible differences between visual and semantic similarity relationships for each category. Stress tests performed on the MDS solutions indicated that the MDS analyses captured meaningful levels of variance in the similarity data. These stimuli, associated norms and naming data are made availableto all researchers, and should provide a useful resource for researchers of vision, memory and conceptual knowledge wishing to run experiments using well-parameterized stimulus sets.

Sinking dynamics and splitting of a granular droplet

A granular droplet that is composed of smaller and denser particles in a bed of larger and lighter particles is found to sink and split when subjected to a combination of vibration and fluidizing gas flow. Despite visual similarities with fluid-like instabilities, the observed phenomenon is a result of the particulate character of granular matter. Combining experiments and numerical simulations, we show that the droplet of high-density particles causes the formation of an immobilized zone that obstructs the downwards motion of the droplet and causes the droplet to spread and ultimately to split. We further investigate the conditions required for droplet splitting.

The Contributions of Letter Features to Arabic Letter Knowledge for Arabic-Speaking Kindergartners

ABSTRACT Studies have suggested that multiple features influence letter knowledge across different orthographies. Arabic offers a unique opportunity to investigate the relations of letter properties on letter knowledge, but research on Arabic letter knowledge is scarce. This study was designed to investigate (a) letter frequency, (b) letter sequence, (c) visual similarity, (d) developmental stage of the phonemes of the letters, and (e) diglossia as possible factors that contribute to Arabic letter knowledge. A total of 142 (Mean age = 67 months) native Arabic-speaking monolingual kindergartners were administered a letter knowledge task. Data were analyzed using Cross-Classified Generalized Random-Effects analysis, which allows partitioning of variance into that due to persons and due to letters to provide a more unbiased estimate of item-level variance. Results showed that letter frequency, late developing sounds, and diglossia were statistically significant when each was entered separately in a model whereas visual similarity and letter sequence were not. However, when all letter features were entered simultaneously, letter frequency was the only feature that contributed significantly to the prediction of Arabic letter knowledge. The present study underscores the important role of letter frequency in Arabic letter knowledge. Implications for assessment and instruction are discussed.