Dongjian He

An Improved Hybrid Model for Automatic Salient Region Detection

In this letter, the graph-based visual saliency (GBVS) model is extended by pulse-coupled neural network (PCNN) to implement the well-defined criteria for a saliency detector. In receptive field, the resized intensity feature map generated by GBVS was regarded as the input image of the PCNN. After modulation, the optimal iteration number and threshold were identified by GBVS and Otsus method in pulse generator part, respectively. Moreover, other parameters of the PCNN were set automatically. In the end, an automatic salient region detection algorithm was proposed. Experimental results show that our proposed hybrid model can efficiently detect salient region.

Ensemble of multiple descriptors for automatic image annotation

Automatic image annotation (AIA) plays an important role and attracts much research attention in image understanding and retrieval. Annotation can be posed as classification problems where each annotation keyword is defined as a group of database images labeled with a semantic word. It is shown that, by establishing one-to-one corresponding between image region and semantic keyword is a feasible approach for automatic image annotation. In this paper, we proposed a novel algorithm, EMDAIA for automatic image annotation based on ensemble of descriptors. EMDAIA regards the annotation process as a multi-class image classification. The producers of EMDAIA are presented as follows. First, each image is segmented into a collection of image regions. For each region, a variety of low-level visual descriptors are extracted. All regions are then clustered into k categories with each cluster associated with an annotation keyword. Moreover, for an unlabeled instance, distance between this instance and each cluster center is measured and the nearest categorys keyword is chosen to annotate it. Experiment results on LabelMe, a benchmark dataset, shows EMDAIA outperforms some recent state-of-the-art automatic image annotation algorithms.

Decision support of farmland intelligent image processing based on multi-inference trees

We meet the basic need of farmland image processing for non-professionals.We realize intelligent selection of image processing steps and common methods.These classified and analyzed based on knowledge description and inference of intelligent decision.We designed the related rules and built the multi-inference trees.We provide decision support for intelligent image processing. In order to meet the basic need of farmland image processing for non-professionals, to realize intelligent selection of image processing steps and common methods which were classified and analyzed based on knowledge description and inference of intelligent decision in this paper. According to the image features and requirements of users, the related rules were designed and then the color spaces, graying methods, de-noising methods, segmentation methods and morphology post-processing methods multi-inference trees were built. A manual evaluation method was adopted to evaluate the artificial and intelligent processing results. Therefore, the selection of image processing steps and methods can be made through these trees, and decision support of intelligent image processing can be realized. The experimental results show that the evaluation ratio of intelligent processing that more than 80 points is up to 83.0%, the average ratio is 75.9%, and the average processing time is about 0.23s per image, which can greatly reduce the workload and blindness of methods selection and achieve higher level of comprehensiveness and operational efficiency.

Blind-Road Location and Recognition in Natural Scene

In this paper, we describe the dual-threshold processing technology based on gray histogram and carry on binary processing to the blind-road gray image according to the real-time requirement of blind man navigation system and the features of blind-road. We achieved binary image filter processing using erosion and dilation improved algorithm of binary morphological approach. An edge location algorithm is present which can determine the location of pixels around the edge and realize the blind-roads localization accurately. Using the value gain from the location algorithm, an edge tracking algorithm is proposed which achieve the blind-roads tracking accurately. The approach proposed in our paper have been fully implemented and tested on real images. The experimental results show that it can not only recognize the blind-road accurately, but retain the obstacle of the blind-road for the following obstacle recognition and other issues which can provide the accurate basis.

Multi-semantic Scene Classification Based on Region of Interest

Automatic semantic scene classification is a challenging research topic in computer vision and it is also a promising solution to scene understanding and image semantic retrieval. In this paper, novel techniques are proposed to implement multi-semantic scene classification. We first extract some regions of interest (ROIs) from each image based on image-driven, bottom-up visual attention model, and then propose two multi-instance multi-label learning algorithms, EMDD-SVM and EMDD-KNN to cope with this problem, where images are viewed as bags, each of which contains a number of instances corresponding to regions of interest and belongs to multiple categories simultaneously. Experimental results show that our ROIs extraction algorithm could obtain different kinds of interested objects effectively under various complex clutters and is highly tolerant to the noise, and that EMDD-SVM and EMDD-KNN algorithms have achieved good performance on multi-semantic scene classification by integrating multi-instance learning and multi-label learning.

Wheat leaf lesion color image segmentation with improved multichannel selection based on the ChanVese model

An improved Chan-Vese(C-V) model for wheat leaf lesion image segmentation is proposed.Methods of adaptive channel selection and weight computation are proposed.The termination criterion greatly reduces the iterations in the level set evolution.Our method shows superior performance of accuracy, efficiency and robustness over others. Because of the characteristics of intensity inhomogeneity, noise, and blurred edges in crop lesion color images, an improved ChanVese (CV) model for wheat leaf lesion segmentation is proposed. First, to make full use of the color information, three color channels are selected from the R, G, B, H, S, and V channels using principal component analysis. In addition, an initial K-means segmentation is used to obtain the initial lesion curve. Second, because the channel weights are artificially determined in the CV model, the ratio of the average object pixel value to average background pixel value is used as the adaptive weight for each of the three channels. Finally, to avoid a long iteration process for the level-set evolution, an efficient termination criterion is presented. The proposed algorithm has the advantages of an adaptive channel selection, adaptive channel weight computation, and very few iterations. Compared with the traditional CV and gradient descent CV (g-CV) models, the proposed segmentation approach has fewer iterations and higher segmentation accuracy.

A New Bi-cubic Triangular Gregory Patch

A basic problem may be often encountered when people construct smooth surface that interpolates the given curve meshes. In general, object surface can be divided into triangular and quadrangular areas by the given curve meshes. The problem is that while the methods satisfy the continuity and interpolation requirements, they often fail to produce pleasing shapes, or in reverse to the condition. In this paper, we propose a new triangular Gregory patch which is a variant of quadrangular Gregory patch and Bezier triangle. It owns more simple mathematical formula and can make G1 continuity more easily controlled.

A novel saliency map extraction method based on improved Itti's model

To obtain a saliency map close to the saliency object as much as possible, an improved bottom-up visual attention model is presented. Firstly, early visual features such as intensity, color and orientation are extracted from an input image at multiple scales; Secondly, three conspicuity maps are created respectively according to early features; Thirdly, three conspicuity maps are combined into a saliency map nonlinearly. In the last step, different from Ittis model, the contribution rate of each conspicuity map to the saliency map is done in inversely proportional to the saliency points area. A set of experiments were carried out to demonstrate the effectiveness of the proposed model. The experimental results show that the algorithm is effective and saliency map accuracy increased by 15–20% compared to Ittis model.

Research on Cow Epidemic Disease Diagnosis Based Certainty Factor

According to actual of cow epidemic disease diagnosis, bidirectional reasoning scheme based on the symptom both characterization and autopsy is proposed, which is based on inexact reasoning of certainty factor. The process of both construction the knowledge database for cow epidemic disease diagnosis and the inexact reasoning based on certainty factor are amplified. A web-based system on the platform of Tomcat+JSP+MySQL+DWR about the cow epidemic disease diagnosis is realized base on MVC. By testing, it is shown that the monitoring 21 kinds of cow epidemic disease in China can be accurately diagnosed, basically. It can be widely used in breeding aquatics village, breeding farm, raising households and so on. It will play a positive role to prevent and control the cow epidemic disease.

Face Recognition Using (2D)^2PCA and Wavelet Packet Decomposition

The correct recognition rate (CRR) and implementation speed are two evaluation criteria for face recognition system. However, it is difficult to boost them when images are taken under different conditions. In this paper, the performance of a recognition method using wavelet packet decomposition (WPD) and two-directional two-dimensional principal component analysis ((2D)2PCA) is explored. First, plot images are obtained via two-level WPD on original image. And then, the feature matrixes of these plot images are extracted using (2D)2PCA. Finally, the method is constructed by fusing the feature matrixes of dasiasuccessfulpsila plot images properly chosen. Experiments on images with different illumination, expressions, and poses from PIE, Yale, and UMIST indicate that the proposed method can get a higher correct recognition rate than performing (2D)2PCA on original image.