semantic classification of images based on their low-level visual features is a challenging task in the field of image retrieval and classification. In this paper, the effect of weighting color, shape and texture feature vectors and also their components on image classification is investigated. The way that the classification rate is affected by the database size is also studied. A database of 1000 images from 10 semantic groups, 100 images in each group, is used. A k-nearest neighbor classifier is employed and the leave-one-out rule is used to evaluate the results. The optimum weights for each type of the feature vectors and also their components are found by a genetic algorithm.  

semantic image segmentation based on Convolutional Neural Networks (CNNs) is one of the main approaches in computer vision area. In convolutional neural network-based approaches, a pre-trained CNN which is trained on the large image classification datasets is generally used as a backend to extract features (image descriptors) from the images. Whereas, the special size of output features from CNN backends are smaller than the input images, by stacking multiple deconvolutional layers to the last layer of backend network, the dimension of output will be the same as the input image. Segmentation using local image descriptors without involving relationships between these local descriptors yield weak and uneven segmentation results. Inspired by these observations, in this research we propose Context-Aware features (CAF) unit. CAF unit generate image-level features using local-image descriptors. This unit can be integrated into different semantic image segmentation architectures. In this study, by adding the proposed CAF unit to the Fully Convolutional Network (FCN) and DeepLab-v3-plus base architectures, the FCN-CAF and DeepLab-v3-plus-CAF architectures are proposed respectively. PASCAL VOC2012 datasets have been used to train the proposed architectures. Experimental results show that the proposed architectures have 2. 7% and 1. 81% accuracy improvement (mIoU) compared to the related basic architectures, respectively.

Nowadays, a large volume of documents is generated daily. These documents generated by different persons, thus, the documents contain spelling errors. Therefore, existence of automatic writing assistance tools such as spell checker/corrector can help to improve their quality. Context-sensitive are misspelled words that have been wrongly converted into another word of the language. Thus, detection of real-word errors requires discourse analysis. In this paper, we propose a language independent discourse-aware discriminative ranker and use information of whole document and a log-linear model for ranking. To evaluate our method, we augment it into two context-sensitive spellchecker systems, one is based on Statistical Machine Translation (SMT) and the other is based on language model. For more evaluation, we also use two different tests. Proposed method causes outperform about 17 %over the SMT base approach with respect to detection and correction recall.

In this study, we evaluate the performance of five semantic search engines available on the web, using 45 criteria in the form of a researcher-made checklist. Criteria provided in the checklist include both common and semantic features. Common criteria or features are those applicable to all search engines and semantic ones are those applicable only to semantic search engines. Findings show that the selected search engines do not have suitable performance and expected efficiency. DuckDuckGo gained the highest points, considering regular features. Cluuz is in the second place with 20 points and Hakia with 18 points was in the third place. Lexxe and Factbites, with scores of 15 and 10 were placed in the next categories according to their points. In semantic features, DuckDuckGo with 10.65 points was in the first place; Hakia with 9.99 points was in the second place, and then the search engines Cluuz with 8.66 Points, Lexxe with 8.65 points and Factbites with 7.32 points were allocated to the next levels. The results also indicated that on the whole, considering ordinary and semantic features, DuckDuckGo with 31.65, Cluuz with 28.66, Hakia with 27.99, Lexxe with 23.65 and Factbites with 17.32 points got the highest scores.

This paper presents a semantic method for aerial image segmentation. Multi-class aerial images are often featured with large intra-class variations and inter-class similarities. Furthermore, shadows, reections and changes in viewpoint, high and varying altitude and variability of natural scene pose serious problems for simultaneous segmentation. The main purpose of segmentation of aerial images is to make subsequent recognition phase straightforward. Present algorithm combines two challenging tasks of segmentation and classification in a manner that no extra recognition phase is needed. This algorithm is supposed to be part of a system which will be developed to automatically locate the appropriate site for Unmanned Aerial Vehicle (UAV) landing. With this perspective, we focused on segregating natural and man-made areas in aerial images. We compared different classifiers and explored the best set of features for this task in an experimental manner. In addition, a certainty based method has been used for integrating color and texture descriptors in a more efficient way. The experimental results over a dataset comprised of 25 high-resolution images show the overall binary segmentation accuracy rate of 91.34%.

Machine translation has been developed over last years. But this technology is still not able to exactly translate texts. Also post-editing the output may takes longer time than the translation process. So having a quality estimation of machine translation output can be very useful. Moreover, Confidence Estimation can be useful for some applications that their goal is to improve machine translation quality such as system combination, regenerating and pruning. But there is not yet any completely satisfactory method for CE task. We propose 5 syntactic and lexico-semantic features that are never used for confidence estimation task. The experimental results show that proposed lexico-semantic feature outperforms the best baseline system (2) by 9.63% in CER, 8.5% in F-measure and 5.1% in negative class F-measure. Moreover the combination of proposed syntactic features outperforms the best baseline system by 4.49% in CER, 4.1% in F-measure and 2% in negative class F-measure.

In recent years, the performance of deep neural networks in improving the image retrieval process has been remarkable. Utilizing deep neural networks; however, leads to poor results in retrieving images with missing regions. The operators’ dysfunctions, who consider the relationship between the image pixels, statistically extract incomplete information from an image, which in turn reduces the number of image features and or leads to features' inaccurate identification. An attempt has been made to eliminate the problem of missing image information through image inpainting techniques; therefore, a content-based image retrieval method is proposed for images with missing regions. In this method, through image inpainting the crucial missing information is reconstructed. The image dataset is being queried to find similar samples. For this purpose, a two-stage inpainting framework based on encoder-decoder is used in the image retrieval system. Also, the features of each image are extracted from the integration and concatenating of content and semantic features. Through using handcraft features such as color and texture image content information is extracted from the Resnet-50 deep neural network. Finally, similar images are retrieved based on the minimum Euclidean distance. The performance of the image retrieval model with missing regions is evaluated with the average precision criterion on the Paris 6K datasets. The best retrieval results are 60.11%, 50.14%, and 42.43% for retrieving the top one, five, and ten samples after reconstructing the image with the most missing regions with a destruction frequency of 6 Hz, respectively.