In this paper, a novel low-level image feature extraction and indexing scheme based on structure-texture image decomposition is presented. The main idea of this work is to decompose database images to structure and texture sub-images to decrease the destructive effects of simultaneous existence of structure and texture information in the image in indexing phase. It is also shown that precision in a typical content-based image retrieval system can considerably increase by combining the feature vectors extracted from structure and texture sub-images. An image database containing 10000 images of 82 different semantic groups is used to evaluate the proposed method. The results confirm the effectiveness of this method.

Spam is one of the problems that has plagued human societies. Although a lot of research has been done in this field, because spammers keep changing their methods like viruses, so there is always a need to provide new solutions in this field. The purpose of the research is to use image texture features to detect image spam. So far, 22 features of image texture have not been used in one place to detect image spam. In this paper, a hybrid method is used to extract key features. In the proposed hybrid method, the co-occurrence matrix of the gray level and chi-square and the threshold of changes in the value of the features are used. The steps mentioned have a great impact on the performance of the categories and improve the accuracy of detection. In the classification stage, the most widely used machine learning algorithms are used to detect image spams, and after obtaining the results of each category, the output of the algorithms used on spam and valid images is examined and compared. The obtained results show that with the help of the proposed method, good detection accuracy can be achieved compared to other methods. Among the reviewed algorithms, the neural network algorithm shows the best performance. The assumed algorithm in other articles shows a lower detection accuracy than the present article, but in the proposed method, it reaches 99.29% detection accuracy.

Texture image analysis is one of the most important working realms of image processing in medical sciences and industry. Up to present, different approaches have been proposed for segmentation of texture images. In this paper, we offered unsupervised texture image segmentation based on Markov Random Field (MRF) model. First, we used Gabor filter with different parameters’ (frequency, orientation) values. The output image of this step clarified different textures and then used low pass Gaussian filter for smoothing the image. These two filters were used as preprocessing stage of texture images. In this research, we used K-means algorithm for initial segmentation. In this study, we used Expectation Maximization (EM) algorithm to estimate parameters, too. Finally, the segmentation was done by Iterated Conditional Modes (ICM) algorithm updating the labels and minimizing the energy function. In order to test the segmentation performance, some of the standard images of Brodatz database are used. The experimental results show the effectiveness of the proposed method.

The watershed transform is a conventional tool for the segmentation of images. Watershed segmentation is often not effective for textured image regions that are perceptually homogeneous. In this paper we describe a new image segmentation algorithm that integrates the measure of spatial variations in texture with the intensity gradients and consists of a number of conceptual stages. In the first stage, texture representation is calculated using vector summation of complex cell responses in different preferred orientations. In the second stage, gradient images are computed for each of the texture features, as well as for grey scale intensity. These gradients are efficiently estimated using a new proposed algorithm based on a hypothesis model of the human visual system. After that, combining these gradient images, a region gradient which highlights the region boundaries is obtained. Watershed transform of the region gradients properly segment the identified regions. Adaptive thresholding on rotational texture features is used to the problem of over segmentation. The combined algorithm produces effective texture and intensity based segmentation for natural and textured images.      

In recent years, facial expression recognition, as an interesting problem in computer vision has been performed by means of static and dynamic methods. Dynamic information plays an important role in recognizing facial expression in the image sequences. However, using the entire dynamic information in the expression image sequences is of higher computational cost compared to the static methods. To reduce the computational cost, instead of entire image sequence, only neutral and emotional faces can be employed. In the previous research, this idea was used by means of Difference of Local Binary Pattern Histogram Sequences (DLBPHS) method in which facial important small displacements were vanished by subtracting Local Binary Pattern (LBP) features of neutral and emotional face images. In this paper, a novel approach is proposed to utilize two face images. In the proposed method, the face component displacements are highlighted by subtracting neutral image from emotional image; then, LBP features are extracted from the difference image as well as the emotional one. Then, the feature vector is created by concatenating two LBP histograms. Finally, a Support Vector Machine (SVM) is used to classify the extracted feature vectors. The proposed method is evaluated on standard databases and the results show a significant accuracy improvement compared to DLBPHS.

Content-based image retrieval is one of the interesting subjects in image processing and machine vision. In image retrieval systems, the query image is compared with images in the database to retrieve images containing similar content. Image comparison is done using features extracted from the query and database images. In this paper, the features are extracted based on the human visual system. Since the human visual system considers the texture and the edge orientation in images for comparison, the colour difference histogram associated with the image’ s texture and edge orientation is extracted as a feature. In this paper, the features are selected using the Shannon entropy criterion. The proposed method is tested using the Corel-5K and Corel-10K databases. The precision and recall criteria were used to evaluate the proposed system. The experimental results show the ability of the proposed system for more accurate retrieval rather than recently content-based image retrieval systems.

The existence of noise in image reduces its quality and hinders analysis of the image. Image noise reduction techniques are often accompanied with artifact, especially in facing with strong noise. Since sensitivity of human visual system is not alike in all areas of image, i. e. smooth and nonsmooth areas, noise removal can be performed considering the textual information of the image. The proposed approach intends to earnestly remove noise from the smooth region as it is more obvious to human visual system. Indeed, the filtered image produces less artifact in nonsoomth region, as the noise is superficially removed from the nonsmooth region. In the proposed method, the image is segmented into smooth and non-smooth regions using entropy information of the image. Then to remove the noise from each region, the diffusion filter with different parameters is used. The proposed method not only removes the noise but also preserves the edges and details of the image. The proposed method was evaluated using several noisy images and images from CSIQ and IVC databases. According to subjective and objective quality results, accuracy of the proposed method in Gaussian noise reduction is better than the previous works.