Search

Computer Vision nowadays uses many Deep Learning techniques in order to make the computer learn data representations from images and image sequences (as in videos). One of the important tasks in this respect is learning the similarity between two given images, which can be readily accomplished by learning a similarity criterion between the images. This can be readily accomplished using Siamese Convolutional Neural Networks (Siamese CNNs). Siamese CNNs can learn a similarity criterion between various kinds of image pairs. The paper presents a survey, which deals with the study of some remarkable papers which have used Siamese CNNs and triplet nets (which are a variation of the Siamese CNNs) in order to learn how similar are two images to one another.

The report suggests the development of design procedures for the synthesis of design solutions for managing product data and design and production procedures at the stages of the life cycle. Variants of the tasks of forming design decisions are given, the requirements for them are formulated. The rules of choosing descriptions of the components of a digital passport and the wording of the similarity criterion are presented, which forms the developed design procedures. Based on the results obtained, an implementation option is proposed for the formation of design decisions.

Dynamic texture similarity ranking is a challenging and still unsolved problem. Evaluation of how well are various dynamic textures similar to humans perception view is extremely difficult even for static textures and requires tedious psychophysical experiments. Human perception principles are largely not understood yet and the dynamic texture perception is further complicated with a distinct way of perceiving spatial and temporal domains, which complicates any similarity criterion definition. We propose a novel dynamic texture criterion based on the Fourier transformation and properties of dynamic texture spatio-temporal frequencies. The presented criterion correlates well with performed psycho-physical tests while maintaining sufficient diversity and descriptiveness.

Internet users constantly make searches to find objects or results of their interest, generally through terms or phrases. Traditional search offers only solutions that take into account just the individual characteristics of the results, and not the relations they have with the rest of the universe. Typically, we are given an ordered list of the results related to the search criterion, which implies the need of changing several times the terms of the query to get to a solution that is more adequate to the intended search goal. As a solution to this problem, Composite Retrieval proposes that the results to a query may be grouped in sets of items (bundles), related through some similarity criterion, but at the same time are complementary. In this work we propose heuristic algorithms for Composite Retrieval which are evaluated experimentally, showing performance improvements over the previous results presented in the literature.

In this research work, a model of teeth recognition has been proposed to identify a person. The persons' teeth image have been matched against the database of teeth images. An algorithm has been developed to recognize the teeth image by using the image processing methods. The author has been set a similarity criterion to match the teeth image against the image database by specifying the threshold value. This similarity criterion is used to identify a person. The labial teeth and gingiva photographic image database (LTG-IDB) have been taken to verify the results. The database consists 90 photographic images from 27 subjects‥ The experiment results has carried on 20 teeth images of same person and 50 teeth images of different persons. MATLAB has been used for this purpose.

A new algorithm is proposed to detect small objects by background suppression. It relies on a modified version of the nonlocal means filter introduced by Buades et al. for background prediction. Background pixels are estimated by a weighted average depending on the similarity between neighborhoods pixels. For background suppression, the similarity criterion is modified to be less sensitive to point object presence. The resulting D-NLM filter better detects dim point objects than current detection filters. A straightforward spatio-temporal extension is also proposed, it has the advantage of not requiring prior accurate background motion compensation. Our method is well adapted to slow object velocity and large background motion contexts, where it outperforms significantly conventional spatio-temporal filters.

This paper presents a family of metrics for assessing image similarity. The methods use the Hilbert-Schmidt Independence Criterion (HSIC) to estimate nonlinear statistical dependence between multidimensional images. The proposed methods have very good theoretical and practical properties. We illustrate the performance in evaluating the quality of natural photographic images, hyperspectral images under different noise levels, in synthetic multiresolution problems, and real pansharpening products.

In integrated condition assessment, family quality is an factor affecting a transformer's condition. If some devices in family have had default record, then the other transformer in family would have same default in future. And now, the affecting degree by family default factor is subjectively decided by expert's experience. This paper collected power transformer experimental data in same factory and with same type, analyzed condition evolution similarity of power transformer in family based on clustering technology to mine the potential evolution rule. To make the clustering result more accurate, this paper improved the similarity criterion in clustering algorithm, proposed line slope distance of condition evolution as line shape similarity criterion, used both data distance criterion and line slope distance criterion to cluster transformer experiment data with same factory and same type in reality. It then analyzed the condition evolution of a power transformer according to the family condition evolution rule. The result is the same with the reality.

In clinical practice, digital subtraction angiography (DSA) is a powerful technique for the visualization of blood vessels in X-ray image sequences. Different with traditional DSA image registration processes, in our proposed image registration method, the control points are selected from the vessel centerlines using multiscale Gabor filters, and mutual information (MI) is then taken as the similarity criterion to find the correspondences. Experimental results demonstrate our algorithm efficiently yields satisfying registration result for DSA images.

We introduce a new technique for nonrigid registration for medical images based on a hierarchical composition of local deformations. The local deformation model is analyzed in order to guarantee continuity, differentiability and a one-to-one transformation by constraining the parameters of the spatial transformation. The performance of our technique was validated with synthetic images, and MR images. The proposed method was always able to improve the similarity criterion, the normalized mutual information, where accuracy was determined strongly by the resolution and dimensions of the involved domains. The searching strategy was carried out with a hybrid genetic algorithm.

The article describes a novel way of representing large databases of shapes. We propose a hierarchical clustering of a set of Fourier-transformed contours. The clustering analysis is density-based and is performed using topographic maps. We have tested the approach on a database of extracted contours of marine animals, generated by F. Mokhtarian et al. (1996). The resulting clusters group contours that show similar global shapes, which sometimes differ from those grouped by Mokhtarian et al., due to a difference in similarity criterion.