Super-resolution for scene text images is a pre-processing of scene text recognition to improve recognition accuracy. This task aims to improve the visual quality of text regions in the images from low-resolution images. Although SR techniques have significantly improved with the recent development of deep learning, it is still challenging to reconstruct high-resolution images for wild images with irregular shapes, severe noise, and blurring. This is because CNN-based methods are based on local calculations and do not consider text-specific characteristics and these are unable to deal with irregular deformations, etc. In this paper, we propose a multi-task learning-based Attentional Feature Fusion Network (MAFF-Net) to reconstruct visually high-quality images from low-resolution images in real scenes. MAFF-Net consists of a reconstruction branch and a super-resolution branch, which are trained simultaneously to share complementary features of the reconstruction model, such as noise reduction and structural information of the text, using the feature representation transfer (FRT) module. In addition, the transformer module, equipped with a 2-D self-attention mechanism, is used to deal with irregular deformations of the text. Then, we attempt to improve the visual quality of the images with severe noise, blurring, and irregular deformations by fusing the attentional features of the different viewpoints of the FRT module and the transformer module, respectively. Experimental results on the benchmark TextZoom dataset show that the proposed method achieves competitive performance with state-of-the-art methods and proves its effectiveness, especially for challenging images.

Imbalance data processing is one of big issues for machine learning. There are some proposed approaches. On the other hand, Maharanobis-Taguchi System (MTS) is a well-known approach in quality engineering. Although the target of both researches are similar, there are few researches combining these methods. In this paper, we focus on the similarity between them and propose a method to handle the imbalance data with MTS. Our proposal makes 2 prediction models, one is based on MTS and the other is based on a machine learning algorithm, from imbalanced data as following steps. First, it divides the training data into the major class, the minor class and the border class by Maharanobis distance gotten by MTS. Secondly, it makes a prediction model from the border class using a machine learning algorithm. This model is the second prediction model. In order to classify new instances, our proposed idea classifies by the first model based on MTS firstly. If it is classified in the major or minor class, the method answers this classification result, otherwise, it classifies by the second models based on a machine learning algorithm. In order to evaluate this idea, we handle some the imbalanced data by our idea and other methods and compare those results. Although the experimental result doesn’t show the advantage of our idea, we get the suggestion for improving our approach.

In this paper, we proposed a modified decision tree learning algorithm. In order to improve the traditional decision tree learning algorithm, we modified a predict phase though exists approached modified a learning phase. Our proposed approach makes a decision tree by a traditional decision tree learning algorithm and predicts new data items' class label by K-NN. The traditional decision tree learning algorithm predicts a class label based on the ratio of class labels in a leaf node. When it is not easy to classify data set according to class labels, leaf nodes includes a lot of data items and class labels. It causes to decrease the accuracy rate. However, it is difficult to prepare good training data set. So we used K-NN to predict a class label from data items in a leaf node. In order to evaluate our approach, we did an experiment using a part of open data sets from UCL learning repository. We compared our approach to ID3 which is one of traditional decision tree learning algorithms and K-NN in this experiment. Experimental result shows our approach is better than ID3 when the leaf nodes include a lot of data items. When the leaf nodes include some data items, our approach can perform like as ID3. So we can say that our approach is useful to modify a decision tree learning algorithm. We don't change a learning process so that our approach doesn't change the readability of a decision tree. In addition to, our approach is better than K-NN. We think that a decision tree works for K-NN as data cleaning. It says that our approach is useful for K-NN. Though we can show the advantage of our approach according to the experiment, there are some data items we can not predict correctly. In future, we have to evaluate experimental results and process in detail. We have to ascertain the cause of error. And we consider how to modify our approach to correct errors. It is likely that normalization is one of useful method. In addition to, we have to evaluate our new approach using some open data sets.

We proposed a modified decision tree learning algorithm to improve this algorithm in this paper. Our proposed approach classifies given data set by a traditional decision tree learning algorithm and cluster analysis and selects whichever is better according to information gain. In order to evaluate our approach, we did an experiment using program-generated data sets. We compared ID3 which is one of well-known decision tree learning algorithm to our approach about the recall ratio in this experiment. Experimental result shows the recall ratio of our approach is similar than the recall ratio of a traditional decision tree learning algorithm. Though we can not show the advantage of our approach according to the experiment, we show it is worth using cluster analysis to make a decision tree. In future, we have to evaluate our approach according to cross-validation method using big and complex data sets in order to say the advantage of our approach. We think our approach is not good for all data set, so we try to find the situation which our approach is better than other approaches according to the experimental results. In addition to, we have to show how to explain a decision tree by our approach to keep the readability of a decision tree.

In this paper, we proposed an idea of improvement of a decision tree learning algorithm using cluster analysis. We classify data set based on two relations. One is the relation between each class and each attribute and the other is the relation between attributes. First relation is used in a traditional decision tree algorithm and second relation is used in cluster analysis. Using second relation is our point in this approach. In order to evaluate our approach, we did an experiment using data set in machine learning repositories. Experimental result show the possibility that our approach is better than a traditional decision tree learning algorithm.

Two types of ontology have been presented to formalize a patient state: mental ontology reflecting the patient mental behavior due to certain disorder and physical ontology reflecting the observed physical behavior exhibited through disorder. The medical knowledge is represented by fuzzy attributes reflected on the medical knowledge. Aggregation function related to physical attributes represented as intuitionistic interval fuzzy weighted ordered average operators. Aggregation functions related to Bonferroni intuitionistic interval fuzzy weighted operators for mental ontology. Fuzzy representation based on these two types of ontology is aligned on medical knowledge for decision making related to medical diagnosis based on pairing function to compute similarity among medical cases. We have constructed an integrated computerized model which reflects a human diagnostician as computer model and through it; an integrated interaction between that model and the real human user (patient) is utilized for 1ststage diagnosis purposes.

In this paper, we propose a personality estimation application for Twitter client in smartphone. This application enables user to know the personality of other Twitter accounts. In this application personality estimation is performed by the text classification method from tweet data. In the demonstration experiment with 44 subjects, our proposed application proves its effectiveness especially for the users in their 20s or Twitter experienced users in entertainment use. Both subjective and objective evaluations show that the personality information in social media makes some contribution to build a new connection and to stimulate the social media use.

Ontology alignment in different view computer Interaction based on emotional modelling and physical views, collectively; has been investigated and reported in this paper. Two types of ontology have been presented to formalize a patient state: mental ontology reflecting the patient mental behavior due to certain disorder and physical ontology reflecting the observed physical behavior exhibited through disorder. These two types of ontology have been mapped and aligned for reasoning using a simple Bayesian Network for causal reasoning to define what we call as simple case diagnosis. We have constructed an integrated computerized model which reflects a human diagnostician as computer model and through it; an integrated interaction between that model and the real human user (patient) is utilized for 1ststage diagnosis purposes.

In this paper, we proposed a method to express emotion in a synthetic speech according to musical knowledge. We define the pitch of each phoneme in speech using an accent dictionary in the first step. Next, we arrange parameters for this synthetic speech to express given emotion in a synthetic speech. Firstly, we arrange tempo, volume and pitch of speech. Next, we arrange pitch of each syllable and connection of syllables. Finally, we arrange pitch of each phoneme according to chord-scale. We decide how to arrange these features according to knowledge from empirical researches about expression emotion in music. In order to evaluate this approach, we did an experiment. The experimental results saw that there was difference between emotion estimated by examinees and emotion tried to express by our approach. However, we could divide emotion in speech into positive and negative. So, we evaluated that the possibility of our approach to express emotion in synthetic speech.

In this paper, we propose a conversational virtual agent application in mobile phone which covers the lack of interactivity of unidirectional private communication. In this application the agent automatically recognize user situation, such as location, locomotion and interruptability using sensors installed in mobile phone. The agent also behaves as if the target person which user wants to contact with behaves based on the Egogram estimation from weblog of this target user. In preliminary experiments, the estimation methods using in this application achieve sufficient accuracy for our purpose at this stage.

Human computer Interaction based on emotional modelling and physical views, collectively; has been investigated and reported in this paper. Two types of ontology have been presented to formalize a patient state: mental ontology reflecting the patient mental behavior due to certain disorder and physical ontology reflecting the observed physical behavior exhibited through disorder. These two types of ontology have been mapped and aligned for reasoning using a simple Bayesian Network for causal reasoning to define what we call as simple case diagnosis. We have constructed an integrated computerized model which reflects a human diagnostician as computer model and through it; an integrated interaction between that model and the real human user (patient) is utilized for 1ststage diagnosis purposes.

The paper proposed an estimation method of human interest/boredom for intelligent HCI system with a single vision system. The system estimates postures of the interactant by the facial feature points acquired from the vision system. The method uses the current software and hardware resources so that no extra device is required. This also results in achieving the real time estimation, i.e., low-cost computation. The experimental result implies that the method has possibilities to evaluate human-interest level as correct as the closer people of the interactant.

In this paper, we proposed a method to express emotion in a synthetic speech using musical theories. We define the pitch of each phoneme according to an accent dictionary in the first step. Next, we change the pitch of each phoneme according to a Chord and a Scale which are good for expression emotion in music. In order to evaluate this approach, we did experiments. The experimental results saw the possibility of using Chords and Scales to express emotion in synthetic speech. Future works of our research are as follows. We have to analysis these synthetic speeches and extract the relation between synthetic speeches and emotion using a machine learning techniques. We have to use other musical theories to express emotion appropriately, in synthetic speech. In addition, we have experimented these results with a lot of listeners to evaluate our approach.

Human computer Interaction based on emotional modeling has been investigated and reported in this paper. Human personality has been analyzed based on ego-gram analysis and accordingly human “SELF” emotional model has been created. We have created as one part a computerized model which reflects a human user (in this paper Medical Doctor model) impeded as a computer model and through it, an emotional interaction between that model and the real human user is established. The interaction scenarios and reasoning are based on transactional analysis. We have implemented the system and empirically, examined it, as experiment in public space for revision and evaluation.

This paper introduces a concept of invariant to estimate the mental states of a person who is interacting with the artifact that has a cloned mentality of the particular person. The invariant in the paper is movements of the facial features that are always observed in the developmental process of the facial expressions expressing the identical mental states. The invariant acquired as the result of cloning the way to judge the mental state of the others by the target of cloning. The interpretation of the facial expression in the paper is done in a subjective manner. An extraction method of the invariants and the mental state estimation with the invariants are described in detail. A brief preliminary experiment shows that the proposed method has a possibility to clone the way by the target person to estimate the mental state from the facial expressions with the invariants.

To implement listener estimate emotion in speech, we propose an approach based on estimation emotion in speaker's speech. First, we use training data consist of speech data and emotion estimated by a listener. We collect human speech data and synthesize speech using speech synthesize technique. Next, we get syllabic features from training data using speech synthesize. We can divide speech into phonemes using speech synthesize. After getting phonemes, we make syllables based on phonemes. We get the fundamental frequency, power and time of each phoneme and calculate the statistics values and the inclination of the regression of them. Next, we make classifiers from these values. Finally, we estimate emotion using them. To evaluate our approach, we did the experiment. The experimental result does not say our approach is strong to do our goal. It shows some points we should modify to enhance our approach. Future works of our research are as follows. We collect training data using speech synthesize. We reconsider speech features for estimation of emotion and make classifiers using them. In addition, we divide emotion into s detail by features before making classifiers. And we evaluate the new approach. We will also modify our approaches to use in real-time.

The paper reports on our experience in adapting emotional experiences of the software engineers in evolutionary design of software systems. The works here reported present development progress report in relation to the state-of-art that need to create the multudisciplinary technologies, needed to establish best harmony engagement between human user the software application, and based on human cognitive analysis. This progress status report outlines the design on what we called as universal template that articulated from the collective experimental data. Several observation participated in the design of what is called universal templates that would be used to interact with human user to articulate on the cognitive model that the user is in, such that to have Kenji System or what is called (certain human mental cloning system) to reason on through mental engagement of the user. In our system we approach the user best engagement from facial and voice analysis. And through it, we can measure (collectivized and quatified), and observe the user behaviour, and accordingly enhance the engagement by generative interactive scenario. The approach has been experimented using famous literature person (Kenji Miyazawa).

This paper describes on a recognition method of facial expression with situational information for virtual person with personality of the existed person. By using situational information, the recognition errors caused by the individual differences in the facial expression are expected to be reduced. As situational information, we adopt profiles of a person such as contour of the face, age, and gender at this time. According to the profiles, we categorize user of the virtual person with whom it shares the time and space for a while, to adopt appropriate template set of facial expressions to correctly recognize the expressions. The paper tries to figure out the virtual person, its interaction with a user, situation, profiles, and the recognition method of facial expressions by using the situational information. We also discuss the reason why the situational information is required to the facial expression recognition method.

To enhance estimation of emotion in speech, we propose three new approaches. First approach is that we use more synthetic speeches than our previous work. We define emotion in these speech based on human evaluation and use these speech data to make classifiers. Second approach is that we add some statistics values to our previous approach. Additional statistics values are quartile, range, interquartile range, the upper and lower half of interquartile range and the coefficient of the regression formula. We assume that these values show new viewpoints about speech features. Third approach is that we use phonemic features and syllabic features to estimate emotion in speech. In this paper, phonemic feature is a feature gotten from each phoneme in a speech by frequency analysis. Syllabic feature is a feature gotten from each syllable in a speech by frequency analysis. We use speech recognition to get phonemes and get syllables from a speech based on phonemes. Experimental result shows phonemic features and syllabic features are more useful than using the fundamental frequency and power to estimate anger, disgust fear and sad. The result also says that additional statistics values hardly contribute to estimate emotion. We need to analysis classifiers to evaluate contribution of these statistics. We have some future works. First work is that we use the frequency and power with phonemic features and syllabic features. Second work is that we modify our approach based on the analysis result of our experiment. Third work is that we use our approach in real-time.

Although Case-Based Reasoning comes out in order to solve knowledge acquisition bottleneck, a case structure acquisition bottleneck has emerged, superseding it. Because we cannot decide an appropriate case structure in advance, a framework for CBR should be able to improve a case structure dynamically, collecting and analyzing cases. Here is discussed a new framework for knowledge acquisition using CBR and model inference. Model Inference tries to obtain new descriptors(predicates) with interaction of a domain expert, regarding the predicate as the slots that compose a case structure, with an eye to the function of theoretical term generation. The framework has two features: (1) CBR obtains a more suitable group of slots (a case structure) incrementally through cooperation with model inference, and (2) model inference with theoretical term capability discovers the rules which deal with a given task better. Furthermore, we evaluate the feasibility of the framework by implementing it to deal with law interpretation and certify two features with the framework.