نشریه ماشین بینایی و پردازش تصویر
سال دوم شماره 1 (بهار و تابستان 1394)

Diffusion Tensor Imaging (DTI) is a common method for the investigation of brain white matter. In this method, it is assumed that diffusion of water molecules is Gaussian and so, it fails in fiber crossings where this assumption does not hold. High Angular Resolution Diffusion Imaging (HARDI) allows more accurate investigation of microstructures of the brain white matter; it can present fiber crossing in each voxel. HARDI contains complex orientation information of the fibers. Therefore, registration of these images is more complicated than the scalar images. In this paper, we propose a HARDI registration algorithm based on the feature vectors that are extracted from the Orientation Distribution Functions (ODFs) in each voxel. Hammer similarity measure is used to match the feature vectors and thin-plate spline (TPS) based registration is used for spatial registration of the skeleton and its neighbors. A re-orientation strategy is utilized to re-orient the ODFs after spatial registration. Finally, we evaluate our method based on the differences in principal diffusion direction and we will show that utilizing the skeleton as landmark in the registration results in accurate alignment of HARDI data.

Content based image retrieval (CBIR) means image retrieval by low level features such as color, texture and shape. In this way, semantic gap is defined as the difference of image interpretation between human and computer algorithm. In this domain, incorrect mapping of low-level features to high-level semantics leads to widening in semantic gap. In image retrieval it is possible to texture, color, and shape of image are changed but the concept is not transitioned in human mind. However, in most cases, feature vector of image is moved in feature space and therefore image is not correctly retrieved. The purpose of this paper is reducing the dimensions of feature vectors by a non-linear approach, learning the manifold space and developing a new feature vector to coincide distances in semantic and feature space domains.So, the continuity between the instances of a semantic at the semantic space is kept in feature space. The main innovation of this paper is extraction of one feature space from multiple ones. In the proposed manner, adverse effect of noise in manifold learning is decreased. Experiments are done on MPEG-7 Part B and Fish datasets and results show effectiveness of proposed methd.

Industrial radiography (RT) is one of the most important non-destructive testing (NDT) methods for the welding defects (e.g. cracks) identification. In the radiography film interpretation, the precise detection of the cracks depends on the interpreter person’ capabilities and skills as well as the image quality of the films.Industrial radiography films often suffer from the low image quality. Therefore, there is a need for accuracy increasing of the defect detection in these type of the radiography images. Cracks have the specific frequency domain feature due to the shape, i.e. very narrow width in comparison with its length. Different frequency and time based image processing techniques can be implemented for the analyzing of the cracks defects. In this research, a wavelet based method and an empirical mode decomposition (EMD) method were used and compared for the detection and analyzing of the welding crack zone. In both methods final image was created as a combination of the original image decomposition components. Then the crack defects have been surveyed in the processed images. The EMD method showed the better results in the crack detection in the comparison of the proposed wavelet method. Although the contrast of the processed images were decreased with EMD method, but also the location and the shape of the cracks clearly appeared.

In this study, a new method is presented for offline signature recognition. To compare the features, a new similarity measure is introduced based on the number of matched features. In addition, in the post-processing step, an innovative, efficient and effective coordinate matching filter is used that has low computational cost and is consistent with the feature extraction algorithm. This filter applies a threshold on Cartesian coordinate difference between the two blocks on the corresponding images. The implementation of the proposed system includes optimized features that are invariant to scaling and rotation changes. Using the new similarity criteria for matching these features, and post-processing routine using the proposed coordinates filter, applied on the GPDS960 (Offline) and SVC2004 (online converted to online) database, improved efficiency of the proposed identification system. Also proposed system parameters are selected and personalized automatically only once by using a genetic algorithm for each database.

Automatic, continuous, non-invasive detection of pain and it’s intensity in clinical settings is needed to assess and manage pain. Therefore, a solution is proposed which deals with the coding of the facial action units associated with pain action based approach to describe the face. Expressions of painvaried in different faces. One of the factors that can create a variety of expression is style of the individual. These expression recognition requires a system that is person independent. In this system, interpersonal variations increase distance between training and test sample distribution, resulting in the loss of generalization of learning. In this paper, in order to reduce the distance between the two distributions, style transfer vector and mapping are proposed to improve pain and its intensity.Results of methods are investigated on the spontaneous UNBC-McMaster database. The results show that style transfer methodsimprove the recognition performance with a sufficient and small amount of adaptation data.

Automatic image annotation refers to automatically assignment of textual labels according to visual content of images. Althoughin the last decade great deal of research has been done in this area, Butbecause of numerous labels and semantic gap between the labels and the low-levelvisualfeatures, the accuracy and efficiency of these systems is reduced. In this study, an annotation method is proposed using two-level clustering based on featureswhich are reduced with genetic algorithm and as well as semantics. Clustering makes visual similar images and also semantic related images be placed next toeach otherandbe annotated. This leads to fast annotation and also has an acceptable performance for an annotation system. To evaluate the proposed method, two well-known datasets, Corel5k and IAPR TC-12 are selected. The results show acceptable performance of the proposed method in comparison with other methods.