Gastrointestinal polyp classification through empirical mode decomposition and neural features

The data interpolation is an essential part of Bidimensional Empirical Mode Decomposition (BEMD) of an image. In the decomposition process, local maxima and minima of the image are extracted at each iteration and then interpolated to form the upper and lower envelopes, respectively. Because of the properties of radial basis function (RBF) interpolators, they are good candidates for use in BEMD. However, only one or two of the RBF interpolators have been utilized for BEMD so far. This paper employs many RBF interpolators for BEMD, compares their performances, and finds out the useful ones for BEMD especially in the image filtering application. We propose to apply the BEMD approach with the adequate interpolation function in the image denoising domain. After that, we combine the BEMD with the DWT to improve the BEMD denoising method. The analysis is done using real images. Simulations are made to focus mainly on the effect of interpolation methods by providing less or negligible control on the other parameters of the BEMD process. The study is believed to work as a guideline in the area of BEMD based real image in the filtering application.

The pansharpening is a combination of multispectral (MS) and panchromatic (PAN) images that produce a high-spatial-spectral-resolution MS images. In multiresolution analysis–based pansharpening schemes, some spatial and spectral distortions are found. It can be reduced by adding spatial detail images of the PAN image into MS images. In the convolution neural network– (CNN) based method, the lowpass filter image extracted by the CNN model when MS and PAN images are directly applied into the input. The feature values are very high and reduce the conversion efficiency. In the proposed scheme, bi-dimensional empirical mode decomposition is used to extract the spatial detail information of the PAN image to reduce the feature values of the input. This extracted PAN image information is applied to the CNN to produce the non-linear changes in the image pixels and transformed into the perfect spatial detail image. It identifies the spatial and spectral detail quantity for the proposed scheme and it also varies with the different datasets automatically of the same satellite images. Simulation results in the context of qualitative and quantitative analysis demonstrate the effectiveness of proposed scheme applied on datasets collected by different satellites.

This paper presents an intelligent system for gastrointestinal polyp detection in endoscopic video. Video endoscopy is a popular diagnostic modality in assessing the gastrointestinal polyps. But the accuracy of diagnosis mostly depends on doctors' experience that is crucial to detect polyps in many cases. Computer‐aided polyp detection is promising to reduce the miss detection rate of polyp and thus improve the accuracy of diagnosis results. The proposed method illustrates an automatic system based on a new color feature extraction scheme as a support for gastrointestinal polyp detection. The scheme is the combination of color empirical mode decomposition features and convolutional neural network features extracted from video frames. The features are fed into a linear support vector machine to train the classifier. Experiments on standard public databases show that the proposed scheme outperforms the previous conventional methods, gaining accuracy of 99.53%, sensitivity of 99.91%, and specificity of 99.15%.

Youzhi Zheng Tsinghua University, Department of Computer Science and Technology, Beijing, China 100084 Zheng Qin Tsinghua University, School of Software, Beijing, China 100084 We propose a novel multiscale decomposition (MSD) image fusion algorithm, which is region-based image fusion using bidimensional empirical mode decomposition (BEMD). BEMD is a 2-D data-driven decomposition derived from the empirical mode decomposition (EMD), which does not require predetermined filter or wavelet function. The input images are decomposed into a number of intrinsic mode functions (IMFs) as well as a residual image. The fusion is performed region by region based on the segmentations of the input images to produce composite BEMD representation, and then the inverse BEMD transform is applied to obtain the fused image. Experiments show that the proposed image fusion algorithm provides superior performance over traditional fusion schemes in terms of both objective metrics and visual quality.

In this paper, we introduce a robust image watermarking method based on bi-dimensional empirical mode decomposition (BEMD) against geometric distortion. Based on the characteristics of the image theory and human visual system, the proposed method makes use of orthogonal properties of empirical mode decomposition to achieve the BEMD transform on the image. The image is decomposed into a series of IMFs and residue which contain the different frequency parts of the image. So the watermark is adaptively weighed to the different positions of the middle frequency IMF region. The method makes use of the multi-scale analysis characteristics of image BEMD theory and the person's sense of vision and shows excellent advantage against shearing attack. The method could show the watermark clearly when half of the image has been cut. Experimental results show this method excellent robustness for image shearing. The watermark thus generated is invisible and performs well in StirMark test and is robust to geometrical attacks. Compared with other watermarking algorithms, this algorithm is more robust, especially against geometric distortion, while having excellent frequency properties.

To better analyze images with the Gaussian white noise, it is necessary to remove the noise before image processing. In this paper, we propose a self-adaptive image denoising method based on bidimensional empirical mode decomposition (BEMD). Firstly, normal probability plot confirms that 2D-IMF of Gaussian white noise images decomposed by BEMD follow the normal distribution. Secondly, energy estimation equation of the ith 2D-IMF (i=2,3,4,......) is proposed referencing that of ith IMF (i=2,3,4,......) obtained by empirical mode decomposition (EMD). Thirdly, the self-adaptive threshold of each 2D-IMF is calculated. Eventually, the algorithm of the self-adaptive image denoising method based on BEMD is described. From the practical perspective, this is applied for denoising of the magnetic resonance images (MRI) of the brain. And the results show it has a better denoising performance compared with other methods.

The interpolation causes the frequency aliasing in intrinsic mode functions (IMF) in bi-dimensional empirical mode decomposition (BEMD). The local orthogonality or the computational complexity of the improved BEMDs can not be met. In this paper, a novel BEMD - local estimation empirical mode decomposition (LEEMD) is proposed. Particularly, the local estimation instead of the interpolation without changing the framework of EMD is proposed to solve the problem. LEEMD will be applied in image fusion based on the proposed background/detail rule. In experiments, the objective and subjective experiments show that LEEMD and its application is effective.

This paper presents a new simple and robust texture analysis feature based on Bidimensional Empirical Mode Decomposition (BEMD) and Local Binary Pattern (LBP). BEMD is a locally adaptive decomposition method and suitable for the analysis of nonlinear or nonstationary signals. Texture images are decomposed to several Bidimensional Intrinsic Mode Functions (BIMFs) by BEMD, which present a new set multi-scale compo- nents of images. In our approach, firstly, saddle points are added as supporting points for interpolation to improve original BEMD, and then images are decomposed by the new BEMD to several components (BIMFs). After then, Local Binary Pattern (LBP) in different sizes is used to detect features from different BIMFs. At last, normalization and BIMFs selection method are adopted for features selection. The proposed feature presents invariant while preserving LBP's simplicity. Our method has also been evaluated in CuRet and KTH-TIPS2a texture image databases. It is experimentally demonstrated that the proposed feature achieves higher classification accuracy than other state-of-the- art texture representation methods, especially in small training samples condition.

The performance of interpolation algorithm used in bi-dimensional empirical mode decomposition directly affects its popularization and application. Therefore, the research on interpolation algorithm is more reasonable, accurate and fast. So far, in the interpolation algorithm adopted by the bi-dimensional empirical mode decomposition, an adaptive interpolation algorithm can be proposed according to the image characteristics. In view of this, this paper proposes an image interpolation algorithm based on the particle swarm and fractal. Its procedure includes: to analyze the given image by using the fractal brown function, to pick up the feature quantity from the image, and then to operate the adaptive image interpolation in terms of the obtained feature quantity. The parameters involved in the interpolation process are optimized by particle swarm optimization algorithm, and the optimal parameters are obtained, which can solve the problem of low efficiency and low precision of interpolation algorithm used in bi-dimensional empirical mode decomposition. It solves the problem that the image cannot be decomposed to obtain accurate and reliable bi-dimensional intrinsic modal function, and realize the fast decomposition of the image. It lays the foundation for the further popularization and application of the bi-dimensional empirical mode decomposition algorithm.

In this paper, we introduce a robust image watermarking method based on bi-dimensional empirical mode decomposition against geometric distortion. Based on the characteristics of the image theory and human visual system, the proposed method makes use of orthogonal properties of empirical mode decomposition to achieve the bi-dimensional empirical mode decomposition transform on the image. The image is decomposed into a series of IMFs and residue which contain the different frequency parts of the image. So the watermark is adaptively weighed to the different positions of the middle frequency IMF region. The method makes use of the multi-scale analysis characteristics of image bi-dimensional empirical mode decomposition theory and the person’s sense of vision and shows excellent advantage against shearing attack. The method could show the watermark clearly when half of the image has been cut. Experimental results show this method excellent robustness for image shearing. The watermark thus generated is invisible and performs well in StirMark test and is robust to geometrical attacks. Compared with other watermarking algorithms, this algorithm is more robust, especially against geometric distortion, while having excellent frequency properties.

This paper proposed a novel multiresolution image fusion approach, which is based on bidimensional empirical mode decomposition (BEMD). The BEMD image is different from the former multi-scale image fusion algorithms which need predetermined filters or wavelet functions. This letter made research on several key issues relevant to the 2D image decomposition and proposed a BEMD image fusion rule to 2D Intrinsic Mode Functions (2D IMF) and residue image of BEMD. The experiments showed the improved performance of this novel algorithm compared with the discrete wavelet transform (DWT).

This paper presents a new method for BEMD. BEMD can decompose a source image into several two-dimensional intrinsic mode functions. During the image decomposition process, it is required to interpolate and draw the upper and lower envelopes. However, these interpolations and drawing the enveloping surface require a large amount of computing time and artificial screening. Thus, some scholars proposed the rapid realization of BEMD. Further, the window size being fixed during the decomposition process led to losses in the data-driven characteristics, adaptability and dimension consistency of the original BEMD. Therefore, this paper proposes a simple but effective means of keeping the original BEMD method features. The estimate reconstruction method is used to replace surface interpolation, and the variable neighborhood window method is adopted to replace the fixed neighborhood window method. In this article, an order filter is used to reconstruct the upper and lower envelopes, and then the filter size is obtained by using the fact that the image information itself is adaptively variable. Through an empirical analysis, this paper shows that this method can keep the original BEMD method’s rapid decomposition, data-driven characteristics, adaptivity and consistency of scale.

Empirical mode decomposition (EMD) based methods have been widely used in fringe pattern analysis, including denoising, detrending, normalization, etc. The common problem of using EMD and Bi-dimensional EMD is the mode mixing problem, which is generally caused by uneven distribution of extrema. In recent years, we have proposed some algorithms to solve the mode mixing problem and further applied these methods in fringe analysis. In this paper, we introduce the development of these methods and show the successful results of two most recent algorithms.

Feature fusion methods have been demonstrated to be effective for many computer vision based applications. These methods generally use multiple hand-crafted features. However, in recent days, features extracted through transfer leaning procedures have been proved to be robust than the hand-crafted features in myriad applications, such as object classification and recognition. The transfer learning is a highly appreciated strategy in deep convolutional neural networks (DCNNs) due to its multifaceted benefits. It heartens us to explore the effect of fusing multiple transfer learning features of different DCNN architectures. Thus, in this work, we extract features of image statistics by exploiting three different pre-trained DCNNs through transfer learning. Then, we transform the features into a generalized subspace through a recently introduced Autoencoder network and fuse them to form intra-class invariant feature vector that is used to train a multi-class Support Vector Machine (SVM). The experimental results on various datasets, including object and action image statistics show that the fusion of multiple transfer learning features improves classification accuracy as compared to fusion of multiple hand-crafted features and usage of single component transfer learning features.

Infrared and visible image fusion based on BEMSD and improved fuzzy set