In Natural Language Understanding, intent detection and slot filling have been widely used to understand user queries. However, current methods tend to rely on single words and sentences to understand complex semantic concepts, and can only consider local information within the sentence. Therefore, they usually cannot capture long-distance dependencies well and are prone to problems where complex intentions in sentences are difficult to recognize. In order to solve the problem of long-distance dependency of the model, this paper uses ConceptNet as an external knowledge source and introduces its extensive semantic information into the multi-intent detection and slot filling model. Specifically, for a certain sentence, based on confidence scores and semantic relationships, the most relevant conceptual knowledge is selected to equip the sentence, and a concept context map with rich information is constructed. Then, the multi-head graph attention mechanism is used to strengthen context correlation and improve the semantic understanding ability of the model. The experimental results indicate that the model has significantly improved performance compared to other models on the MixATIS and MixSNIPS multi-intent datasets.

Chinese Named Entity Recognition is the fundamental technology in the field of the Chinese Natural Language Process. It is extensively adopted into information extraction, intelligent question answering, and knowledge graph. Nevertheless, due to the diversity and complexity of Chinese, most Chinese NER methods fail to sufficiently capture the character granularity semantics, which affects the performance of the Chinese NER. In this work, we propose DSKE-Chinese NER: Chinese Named Entity Recognition based on Dictionary Semantic Knowledge Enhancement. We novelly integrate the semantic information of character granularity into the vector space of characters and acquire the vector representation containing semantic information by the attention mechanism. In addition, we verify the appropriate number of semantic layers through the comparative experiment. Experiments on public Chinese datasets such as Weibo, Resume and MSRA show that the model outperforms character-based LSTM baselines.

Machine reading comprehension with multi-hop reasoning always suffers from reasoning path breaking due to the lack of world knowledge, which always results in wrong answer detection. In this paper, we analyze what knowledge the previous work lacks, e.g., dependency relations and commonsense. Based on our analysis, we propose a Multi-dimensional Knowledge enhanced Graph Network, named MKGN, which exploits specific knowledge to repair the knowledge gap in reasoning process. Specifically, our approach incorporates not only entities and dependency relations through various graph neural networks, but also commonsense knowledge by a bidirectional attention mechanism, which aims to enhance representations of both question and contexts. Besides, to make the most of multi-dimensional knowledge, we investigate two kinds of fusion architectures, i.e., in the sequential and parallel manner. Experimental results on HotpotQA dataset demonstrate the effectiveness of our approach and verify that using multi-dimensional knowledge, especially dependency relations and commonsense, can indeed improve the reasoning process and contribute to correct answer detection.

The eyes are continuously fluctuating during fixation. These fluctuations are called fixational eye movements. Fixational eye movements consist of tremors, microsaccades, and ocular drifts. Fixational eye movements aid our vision by shaping spatial-temporal characteristics. Here, it is known that photoreceptors, the first input layer of the retinal network, have a spatially non-uniform cell alignment called the cone mosaic. The roles of fixational eye movements are being gradually uncovered; however, the effects of the cone mosaic are not considered. Here we constructed a large-scale visual system model to explore the effect of the cone mosaic on the visual signal processing associated with fixational eye movements. The visual system model consisted of a brainstem, eye optics, and photoreceptors. In the simulation, we focused on the roles of fixational eye movements on signal processing with sparse sampling by photoreceptors given their spatially non-uniform mosaic. To analyze quantitatively the effect of fixational eye movements, the capacity of information processing in the simulated photoreceptor responses was evaluated by information rate. We confirmed that the information rate by sparse sampling due to the cone mosaic was increased with fixational eye movements. We also confirmed that the increase of the information rate was derived from the increase of the responses for the edges of objects. These results suggest that visual information is already enhanced at the level of the photoreceptors by fixational eye movements.

A band-pass bilateral filter is an improved variant of a bilateral filter that does not have low-pass characteristics but has band-pass characteristics. Unfortunately, its computation time is relatively large since all pixels are subjected to Gaussian calculation. To solve this problem, we pay attention to a nonlinear filter called ε-filter and propose an advanced ε-filter labeled band-pass ε-filter. As ε-filter has low-pass characteristics due to spatial filtering, it does not enhance the image contrast. On the other hand, band-pass ε-filter does not have low-pass characteristics but has band-pass characteristics to enhance the image contrast around edges unlike ε-filter. The filter works not only as a noise reduction filter but also as an edge detection filter depending on the filter setting. Due to its simple design, the calculation cost is relatively small compared to the band-pass bilateral filter. To show the effectiveness of the proposed method, we report the results of some comparison experiments on the filter characteristics and computational cost.

Digital halftoning is used to quantize a grayscale image to a binary image. Error diffusion halftoning generates a high-quality binary image, but also generates some defects such as the warm effect, sharpening, and so forth. To reduce these defects, Kite proposed a modified threshold modulation method that utilizes a multiplicative parameter for controlling sharpening. Nevertheless, some degradation was observed near the edges of objects with a large luminance change. In this paper, we propose a method of controlling the multiplicative parameter in proportion to the magnitude of the local edge slope. The results of computer simulation show a greater reduction of sharpening in the halftone image. In particular, there is a great improvement in the quality of the edges of objects with a large luminance change.

Edge enhancement of noisy ultrasound images is important for medical diagnosis. Conventional edge enhancement methods are mainly directed to emphasizing the high-frequency components of the image. Because these methods emphasize also the noise of image, they are not suitable for noisy ultrasound images with speckle noise. In this paper, we propose an edge enhancement method using mathematical morphology based on a geometrical characteristics of the image, using locally variable structuring elements. We show that the proposed method enhances the edge of ultrasound images without noise emphasis.

This paper propose a method for improving the image quality of motion estimation (ME) using low-bit images. By using edge-enhanced images for quantization, we can increase the accuracy of the ME and improve the image quality. It is known that using low-bit images for ME is effective for reducing power consumption but it slightly degrades image quality. The quality of the encoded image depends on the thresholds for data quantization, thus, algorithms for determining thresholds are studied. The proposed method uses linear quantization, which simply truncates the least significant bits. This method is simple without any complicated threshold calculations, and the resultant image quality is improved as much as the methods that use threshold calculations. To evaluate the effectiveness, we simulate results for image quality and estimate the power consumption using synthesis results from a VHDL model motion estimator.