Quaternion wavelet transform (QWT) combines discrete wavelet transform (DWT) and quaternion Fourier transform (QFT). QWT has many applications included imageprocessing. In this research, we discuss about construction...
Quaternion wavelet transform (QWT) combines discrete wavelet transform (DWT) and quaternion Fourier transform (QFT). QWT has many applications included imageprocessing. In this research, we discuss about construction, characteristics and implementation of QWT on process of image denoising. We construct denoising algorithm with QWT then we do simulation to know performance of algorithm. We use grayscale test images that have size 512 × 512 pixel with low, medium and high complexity. Experiment removes noise of image successfully. Results of image denoising are used to measure algorithm performance using PSNR (peak signal to noise ratio) value. We compare PSNR values with DWT and QWT for Haar, Biorthogonal, Daubechies and Coiflets wavelet. The method that has the highest PSNR value can be concluded the best performance.
Convolutional Neural Networks (CNN's) are known to perform well on computer vision tasks such as image classification, image segmentation, and object detection. However, one major drawback of CNN's is the huge...
Convolutional Neural Networks (CNN's) are known to perform well on computer vision tasks such as image classification, image segmentation, and object detection. However, one major drawback of CNN's is the huge amount of computing and memory resources needed to train them. In this paper, we propose an architectural unit which we call Upsampling-Based wavelet Residual Block (UBWRB), that utilizes the 2D discrete wavelet transform coupled with upsampling operators and a residual connection to extract features from image data while having relatively fewer trainable parameters as compared to traditional convolutional layers. The discrete wavelet transform is a family of transforms that find extensive applications in signalprocessing and time-frequency analysis. For this paper, we use the filter-bank implementation of the discrete wavelet transform, allowing it to act in a similar fashion to a convolutional layer with fixed kernel weights. We demonstrate the performance and parameter-efficiency of CNN's with UBWRB's in the task of image classification by training them on the MNIST, Fashion-MNIST, and CIFAR-10 datasets. Our best-performing models achieve a test accuracy of 99.34% on the MNIST dataset while having less than 120,000 trainable parameters, and 92.90% and 84.27% on the Fashion-MNIST and CIFAR-10 datasets respectively, with both having less than 180,000 trainable parameters.
The article presents the results of studies on interference suppression in optoelectronic methods for monitoring weft thread weaving looms on the basis of linear image sensors caused by inhomogeneities of the backgrou...
The article presents the results of studies on interference suppression in optoelectronic methods for monitoring weft thread weaving looms on the basis of linear image sensors caused by inhomogeneities of the background created by sources of natural or artificial light sources in a controlled image scene. Existing solutions that use linear or nonlinear filtering are focused primarily on the processing of two-dimensional images on personal computers and are unsuitable for use in embedded applications. In particular, their use is also ineffective for solving the above problem, when the useful signal is defined as the difference between the output signals of the photodetector taken at a fixed time interval. As studies have shown, in this case, the best result is the use of discrete wavelet-transform.
The Quality of vision is a practical objective in the image *** of the applications of imageprocessing procedures is image Fusion. To get the subjective vision of a image by gathering the best data from source images...
The Quality of vision is a practical objective in the image *** of the applications of imageprocessing procedures is image Fusion. To get the subjective vision of a image by gathering the best data from source images of a similar scene/picture and spot them in a solitary void image is the image Fusion process. A simple and versatile approaches are statistical measures, that are applicable in any kind of image/signalprocessing techniques. In the first step, image is decomposed using Discrete wavelet Transform (DWT). Mathematical computation of Smoothness is proposed in transform domain. This metric is used to select important information from multiple source images resulting to fused image. Further, Human Visual System (HVS) is also explored for fusion. Here all sub bands of DWT are multiplied with HVS weights. Highest response sub band is identified from various sub bands of multiple images using HVS. These sub bands are selected to get the fused image. Smoothness based fusion technique identifies the good texture information and leave the noise affected portions from the multiple source images. HVS based fusion technique identifies the visually important information from the multiple source images. The registered multi-focus and medical images are considered as source images. The experimental results shows that proposed fusion techniques are good in terms of popular fusion metrics.
暂无评论