Comparison of filter banks of DWT in recovery of image using one dimensional signal vector

In recent years, with the rapid development of science and technology, people’s demand for the quality of daily production and life has gradually increased, and all localities are gradually urbanized. The convenient use of water conservancy, transportation, environment, medical treatment, power grid and other aspects of cities in any country affects people’s healthy life. Therefore, it is urgent to build a smart city. For smart city, intelligent identification and management is a very large and complex problem. As one of the high and new technologies, the intelligent recognition of machine vision derived from artificial intelligence has always been a hot spot of great concern. It just provides convenience for the development of smart city and becomes the development direction of smart city construction in the future. Based on this, the role of machine vision is to connect with the computer through wireless sensors such as cameras to simulate an eye that can represent human visual function. This simulated eye can be recognized intelligently in real time. It transmits the recognized information to the computer, and the computer will analyze and process the obtained information for judgment and recognition. This paper will mainly use the optimal threshold segmentation algorithm based on Machine Vision video processing to solve the problem of urban intelligent recognition, and use a specific algorithm to solve some difficulties and obstacles in intelligent recognition. The traditional optimal threshold segmentation algorithm, the optimal threshold segmentation algorithm and the improved optimal threshold segmentation algorithm are experimentally compared. After experimental comparison, it is found that the optimal threshold segmentation algorithm, the improved optimal threshold segmentation algorithm and the traditional optimal threshold segmentation algorithm can intelligently identify the types of buildings and urban traffic signs in the city, Compared with the improved optimal threshold segmentation algorithm, the improved optimal threshold segmentation algorithm improves the response speed, real-time performance, stability and accuracy of the algorithm. Therefore, the optimal threshold segmentation algorithm after the well meets the needs of building the system, and can be useful in the process of intelligent recognition. This improvement is also necessary, which is conducive to the subsequent system construction. 

/pdf/research-on-intelligent-recognition-and-management-of-smart-3navt92c.pdf

Research on Intelligent Recognition and Management of Smart City Based on Machine Vision

The GHM multi-level discrete wavelet transform is proposed as preprocessing for image super resolution with convolutional neural networks. Previous works perform analysis with the Haar wavelet only. In this work, 37 single-level wavelets are experimentally analyzed from Haar, Daubechies, Biorthogonal, Reverse Biorthogonal, Coiﬂets, and Symlets wavelet families. All single-level wavelets report similar results indicating that the convolutional neural network is invariant to choice of wavelet in a single-level ﬁlter approach. However, the GHM multi-level wavelet achieves higher quality reconstructions than the single-level wavelets. Three large data sets are used for the experiments: DIV2K, a dataset of textures, and a dataset of satellite images. The approximate high resolution images are compared using seven objective error measurements. A convolutional neural network based approach using wavelet transformed images has good results in the literature. the effects of 37 different single-level wavelet transformations of images used as ﬁltering for convolutional neural networks to achieve single image super resolution. The convolutional neural networks predict the difference between the coefﬁcients of the wavelet decomposition for the input image and the high-resolution ground truth. This proposes the use of the GHM multi-level wavelet as a superior image representation to achieve deep image super resolution. The GHM wavelet outperforms single-level wavelet representations, while the single-level wavelet representations behave similarly. indicates the structure of the representation is more important than how well the ﬁlter responses maintain properties of the image important to the human-visual system such as edge texture

GHM Wavelet Transform for Deep Image Super Resolution

The accurate and automatic reconstruction of a super-resolution image from one single low-resolution image is an important challenging problem for image analysis. In this paper, we have optimized the problems of existing methods and presented an automated super-resolution image reconstruction system from a single low-resolution image based on cascaded recursive residual convolutional neural network. In this method, we extracted the characteristics of images firstly, then used recursive networks to connect the feature information of each layer to obtain three residual images. Finally, the high-resolution image was obtained by applying the deconvolution at the end of network. we used the recursive structure applying to the residual learning, this network differed from traditional residual learning in that the structure and overlay of the remaining learning network are different, it can solved the problem of gradient explosion, overfitting and insufficient transmission of high frequency component. Compared with present four methods, our algorithm had good performance in experiment of super-resolution testing. Experimental results showed that the proposed method could recover the fine details and textures, simultaneously, it has satisfactory performance in PSNR values.

Single Image Super-Resolution Based on Cascaded Recursive Residual Convolutional Neural Network

The obtained precisely high frequency information is the key of single-frame image super-resolution reconstruction by using two-dimensional wavelet. Because the bicubic interpolation of high frequency components decomposed by wavelet will introduce noise, it will affect reconstruction effect. An improved algorithm using Fourier transform and zero-padding resampling instead of bicubic interpolation is proposed in this paper. The advantage of frequency domain interpolation is obtained by using Fourier transform and zero-padding resampling. And high frequency components obtained by wavelet decomposition of the original image can be interpolated optimally without introducing noise, which makes the high frequency details more precise in the reconstruction process. The experimental results show that the improved algorithm is superior to the traditional two-dimensional wavelet reconstruction algorithm, which can be applied to the single-frame remote sensing image super-resolution reconstruction.

Jinghong Zhou

Papers

Single-Frame Remote Sensing Image Super-Resolution Reconstruction Algorithm Based on Two-Dimensional Wavelet