Today, with the development of technologies to capture and share images, the number of digital images has increased significantly. The management of this volume of images requires an efficient system to review, classify, search and retrieve the images. New generations of image retrieval systems usually take one or a few keywords from the user to retrieve images with visual content related to that keywords. A mechanism that can automatically describe the content of an image (like a human) can increase the efficiency of these systems. automatic image annotation or AIA is a professional method to express the content of images by keywords or tags. AIA systems, investigate the relationship between the meaning of a text and low-level image features by using machine learning techniques. They automatically assign some tags to images to facilitate fast search based on image contents. In this paper, we explain the different steps to implement an AIA system, review the related works and express the problems and challenges in designing such systems. We also introduce several datasets suitable for AIA systems.

image annotation systems are responsible for describing the content of the images by assigning tags to them. The purpose of this research is to improve the accuracy and speed of image annotation system. Recently, with the growing of images, the image annotation process is based on the basics of images instead of themselves. One of these new methods is the implementation of the non-negative matrix algorithm (NMF) on the features of the images. In the proposed method, for the first time, in order to increase the speed and efficiency of the7 system, we use a method that called the block principal pivoting for the NMF solution. This method has ability to add online new class of data to its knowledge and knowledge learning in a compact form. Moreover, the ability to train based on received data without having to be re-processed. In the training phase, the matrix of the coefficients and the base of the input images are obtained using the Block Principal Pivoting method. Then, at the test phase for the input image, by extracted features of the image and the coefficients obtained from the training phase, the coefficient of belonging to the test image is calculated to each of the classes of training images. Then, this coefficient while searching among the teaching images for assigning the label to test image increases the accuracy of the algorithm. This search is done by the KNN method on the base of the images. To test the proposed method, we used two databases Corel5K and real animal data (derived from 500px) and, finally, compared with existing methods, which we found in the Corel5K database at a precision of 50. 20 and real data was 62. 89. Precision have been increased considerably.

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.

automatic image annotation is a process in which computer systems automatically assign the textual tags related with visual content to a query image. In most cases, inappropriate tags generated by the users as well as the images without any tags among the challenges available in this field have a negative effect on the query's result. In this paper, a new method is presented for automatic image annotation with the aim at improving the obtained tags, as well as reducing the effect of unrelated tags. In the proposed method, first, the initial tags are determined by extracting the low-level features of the image and using neighbor voting method. Afterwards, each initial tag is assigned by a degree based on the neighbor image features of the query image. Finally, they will be ranked based on summing the degrees of each tag and the best tags will be selected by removing the unrelated tags. The experiments conducted on the proposed method using the NUSWIDE dataset and the commonly used evaluation metrics demonstrate the effectiveness of the proposed system compared to the previous works.

Graph based semi-supervised methods for automatic image annotation are mainly focused on single-label problems. However, most of the real world problems require multiple labels per image. As a hybrid semi-supervised approach, LGC+ML-KNN is proposed for multi-label image annotation. LGC is a graph based semi-supervised learning algorithm that annotates unlabeled samples. Subsequently, ML-KNN learns from many more labeled samples, as compared to the initial training set. Experiments on several datasets confirm that the proposed approach has better accuracy than available methods, especially when a very small portion of the training set are the labeled samples.

Since most of visual data is stored in the compressed form, investigating semantic retrieval techniques with the description capability of image semantics in the image compression domain is highly desirable. Regardless of the fact that content based image retrieval (CBIR) based on the Vector Quantization (VQ) compression method is more accurate than the other methods, it is expected that semantic retrieval can also be effective. Thus, the goal of this study is to develop a novel automatic image annotation method in the compressed domain. To this end, firstly the images are compressed using the VQ compression method and then are segmented into equal rectangular regions. Each region in the labelled image will be assigned a visual weight that will be calculated. In the annotation process, the relevance model which is a joint probability distribution of the word annotations and the image regional and global features vector is computed through the training set. Therefore, the unlabelled images are annotated. Finally, the image is retrieved on the basis of its semantic concepts. The experiments over 5k Corel images have shown that the retrieval performance of the method suggested here is higher than that of other methods in the uncompressed domain.

By increasing the number of images, it is essential to provide fast search methods and intelligent filtering of images. To handle images in large datasets, some relevant tags are assigned to each image to for describing its content. automatic image annotation (AIA) aims to automatically assign a group of keywords to an image based on visual content of the image. AIA frameworks have two main stages; Feature Extraction and Tag Assignment which are both important in order to reach a proper performance. In the first stage of our proposed method, we utilize deep models to obtain a visual representation of images. We apply different pre-trained architectures of Convolutional Neural Networks (CNN) to the input image including Vgg16, Dense169, and ResNet 101. After passing the image through the layers of CNN, we obtain a single feature vector from the layer before the last layer, resulting into a rich representation for the visual content of the image. One advantage of deep feature extractor is that it substitutes a single feature vector instead of multiple feature vectors and thus, there is no need for combining multiple features. In the second stage, some tags are assigned from training images to a test image which is called “ Tag Assignment” . Our approach for image annotation belongs to the search-based methods which have high performance in spite of simple structure. Although it is even more time-consuming due to its method of comparing the test image to every training in order to find similar images. Despite the efficiency of automatic image annotation methods, it is challenging to provide a scalable method for large-scale datasets. In this paper, to solve this challenge, we propose a novel approach to summarize training database (images and their relevant tags) into a small number of prototypes. To this end, we apply a clustering algorithm on the visual descriptors of training images to extract the visual part of prototypes. Since the number of clusters is much smaller than the number of images, a good level of summarization will be achieved using our approach. In the next step, we extract the labels of prototypes based on the labels of input images in the dataset. because of this, semantic labels are propagated from training images to the prototypes using a label propagation process on a graph. In this graph, there is one node for each input image and one node for each prototypes. This means that we have a graph with union of input images and prototypes. Then, to extract the edges of graph, the visual feature of each node on graph is coded using other nodes to obtain its K-nearest neighbors. This goal is achieved by using Locality-constraints Linear Coding algorithm. After construction the above graph, a label propagation algorithm is applied on the graph to extract the labels of prototypes. Based on this approach, we achieve a set of labeled prototypes which can be used for annotating every test image. To assign tags for an input image, we propose an adaptive thresholding method that finds the labels of a new image using a linear interpolation from the labels of learned prototypes. The proposed method can reduce the size of a training dataset to 22. 6% of its original size. This issue will considerably reduce the annotation time such that, compared to the state-of-the-art search-based methods such as 2PKNN, the proposed method is at least 4. 2 times faster than 2PKNN, while the performance of annotation process in terms of Precision, Recall and F1 will be maintained on different datasets.

In this paper, a method is proposed to automatically select reference image in histogram matching. Histogram matching is one of the simplest spatial image enhancement methods which improves contrast of the initial image based on histogram of the reference image. In the conventional histogram matching methods, user should perform several experiments on various images to find a suitable reference image. This paper presents a new method to automatically select the reference image. In this method, images are converted from RGB to HSV, and the illumination (V) components are considered to select the reference image. The appropriate reference image is selected using a similarity measure via measuring the similarity between the histograms of the initial image and histograms of the images in the data base. Indeed, an image with similar histogram to the histogram of the original images is more appropriate to choose as the reference image for histogram matching. Results in this research indicate superiority of the proposed approach, compared to other existing approaches, in image enhancement via histogram matching. In addition, the user would have no concern in selecting an appropriate reference image for histogram matching in the proposed approach. This approach is applicable to both RGB and gray scale images.