Sharif Digital Repository

Automatic image annotation methods generate a list of tags for each test image and present it in a matrix structure. To achieve a more accurate annotation, we propose a method with the aim of correcting the tag list. In our method, we detect an indicator for each group of tags and use it to rectify the annotation results. To find a correct indicator, we apply a deep feature vector generated by the “AlexNet” model. Using this indicator, we determine the suitable tags for an image. The purposed method is independent of feature vector, dataset, and annotation method. It can be applied to the currently available annotation methods. Our experiments showed improvement in all annotation methods... 

The goal of this paper was to create a new method for analyzing the online real estate flyers based on their property types. We created an algorithm which identifies the buildings and windows from the buildings in order to extract some useful features for classifying the flyers. Our novel approach for building recognition has two main steps: 1- Building Detector 2- Region Growing. Our novel window detection algorithm uses vanishing point to identify nearly the best angle for applying window detection. It transforms the 2D image into 3D and rotates the 3D image around the z-axis and picks the appropriate angle based on the vanishing points. Using these two novel techniques we were be able to... 

Feature vector extraction, based on local image texture, is a primitive algorithm for many other applications, like segmentation, clustering and identification. If these feature vectors are a good match to the human visual system (HVS), we can expect to get the appropriate results by using them. Gabor filters has been used for this purpose successfully. In this paper we introduce a novel refinement, with the use of Matching Pursuit (MP) to improve the Gabor based texture feature extractor. With this improvement, we show that the separability of different textures will increase. Another consideration in this work is computation complexity. Therefore, we limit the basis function set to reduce... 

This work presents a method to increase the face recognition accuracy using a combination of Wavelet, PCA, KPCA, and RBF Neural Networks. Preprocessing, feature extraction and classification rules are three crucial issues for face recognition. This paper presents a hybrid approach to employ these issues. For preprocessing and feature extraction steps, we apply a combination of wavelet transform, PCA and KPCA. During the classification stage, the Neural Network (RBF) is explored to achieve a robust decision in presence of wide facial variations. At first derives a feature vector from a set of downsampled wavelet representation of face images, then the resulting PCA-based linear features and... 

We propose a blind image steganalysis method based on the Singular Value Decomposition (SVD) of the Discrete Cosine Transform (DCT) coefficients that are revisited in this work. We compute geometric mean, mean of log values, and statistical moments (mean, variance and skewness) of the SVDs of the DCT sub-blocks that are averaged over the whole image to construct a 480-element feature vector for steganalysis. These features are fed to a Support Vector Machine (SVM) classifier to discriminate between stego and cover images. Experimental results show that the proposed method outperforms most powerful steganalyzers when applied to some well-known steganography algorithms  

In feature-based image registration, feature selection and reduction methods play an important role in decreasing computational burden of these operations. In this paper, a new approach is introduced to reduce the dimension of extracted feature vectors of MR images. This approach is based on the selection of the maximum and minimum responses of the alpha stable filter for the MR images over the extracted features with different orientation in frequency domain. This algorithm selects the rotation invariant features which are suitable for image registration purposes. It has been shown that these features could efficiently describe the image elements. The discriminating ability of the features... 

This paper addresses a new approach to the keyframe extraction problem employing generalized Gaussian density (GGD) parameters of wavelet transform subbands along with Kullback-Leibler distance (KLD) measurement. Shot and cluster boundaries are selected using KLDs between GGD feature vectors, and then keyframes are located based on similarity and dissimilarity criteria. Objective and subjective evaluations show the high accuracy of this new approach compared with traditional methods  

3D scene understanding is an open challenge in the field of computer vision. Most of the focus is on 2D methods in which the semantic labeling of each RGB pixel is considered. But, in this paper, the 3D semantic labeling of RGB-D images is considered. In the proposed method, to extract some meaningful features, the superpixel generation algorithm is applied to the RGB image to segment it into a set of disjoint pixels. After that, the set of three powerful classifiers are utilized to semantically label each superpixel. In the proposed method, the probability outputs of these classifiers are concatenated as the novel feature vector for each superpixel. Consequently, to analyze the strengths... 

In this paper a new method for automatic word clustering is presented. We used this method for building n-gram language models for Persian continuous speech recognition (CSR) systems. In this method, each word is specified by a feature vector that represents the statistics of parts of speech (POS) of that word. The feature vectors are clustered by k-means algorithm. Using this method causes a reduction in time complexity which is a defect in other automatic clustering methods. Also, the problem of high perplexity in manual clustering methods is abated. The experimental results are based on "Persian Text Corpus" which contains about 9 million words. The extracted language models are evaluated... 

Shape analysis plays an important role in many medical imaging studies. One of the recent shape analysis methods uses the Laplace Beltrami operator which is also used in this paper for hippocampal shape comparison. We proposed a feature vector which consists of size measures and shape descriptors based on Laplace Beltrami eigenvalues and eigenfunctions. The aforementioned feature space is utilised for automatic differentiating normal subjects from epileptic patients as well as distinguishing epileptic patients with left temporal lobe epilepsy (LTLE) from patients with right temporal lobe epilepsy (RTLE). Achieved results are diagnostic accuracy of 93% with 95% sensitivity and lateralization... 

A classifier based on Support Vector Data Description (SVDD) is proposed for spoken digit recognition. We use the Mel Frequency Discrete Wavelet Coefficients (MFDWC) and the Mel Frequency cepstral Coefficients (MFCC) as the feature vectors. The proposed classifier is compared to the HMM and results are promising and we show the HMM and SVDD classifiers have equal accuracy rates. The performance of the proposed features and SVDD classifier with several kernel functions are evaluated and compared in clean and noisy speech. Because of multi resolution and localization of the Wavelet Transform (WT) and using SVDD, experiments on the spoken digit recognition systems based on MFDWC features and... 

In this paper we use a feature vector consisting of the Mel Frequency Discrete Wavelet Coefficients to recognize spoken phonemes in the Persian language. The purpose of using wavelet in feature extraction is to benefit from its multi resolution analysis and localization property in time and frequency domains. The MFDWCs are obtained by applying the Discrete Wavelet Transform (DWT) to the Mel-scaled log filter bank energies of a speech frame. Feature vectors are used for the HMM-based phoneme recognition on a portion of the FarsDat Persian language database consisting of 35 hour recorded data for training and 15 hour for testing. We evaluate the performance of new features for clean speech... 

Recently many works focus on the vehicle type recognition because it is important in security and authentication systems. Computational complexity and low recognition rate especially when the system has to recognize among a large number of vehicles, are two major problems in vehicle type recognition. In recent years wavelet and contourlet transform have been applied in the recognition tasks successfully. In this paper we proposed a method for recognizing vehicle type in different lighting conditions. We used wavelet and contourlet as tools for feature extraction. These features are powerful and robust to illumination and scale variation. We reduced the dimension of feature vector by resizing... 

In this paper an off-line ArabiclFarsi handwritten recognition Algorithm on a subset of Farsi name is proposed. In this system, There is no sub-word segmentation phase. Script database includes 3300 images of 30 Farsi common names. The features are wavelet coefficients extracted from smoothed word image profiles in four standard directions. The Centers of competitive layer of RBF neural network have been determined by combining GA and K-Means clustering algorithm. Weights of supervised layer has been trained by using LMS rule and the distances of feature vector of each sample to the centre of RBF network have been computed based on warping function. Experimental results show advantages of... 

A combined texture- and color-based skin detection is proposed in this paper. Nonsubsampled contourlet transform is used to represent texture of the whole image. Local neighbor contourlet coefficients of a pixel are used as feature vectors to classify each pixel. Dimensionality reduction is addressed through principal component analysis (PCA) to remedy the curse of dimensionality in the training phase. Before texture classification, the pixel is tested to determine whether it is skin-colored. Therefore, the classifier is learned to discriminate skin and non-skin texture for skin colored regions. A multi-layer perceptron is then trained using the feature vectors in the PCA reduced space. The... 

Class-based n-gram language models are those most frequently-used in continuous speech recognition systems, especially for languages for which no richly annotated corpora are available. Various word clustering algorithms have been proposed to build such class-based models. In this work, we discuss the superiority of soft approaches to class construction, whereby each word can be assigned to more than one class. We also propose a new method for possibilistic word clustering. The possibilistic C-mean algorithm is used as our clustering method. Various parameters of this algorithm are investigated; e.g., centroid initialization, distance measure, and words' feature vector. In the experiments...