Deep learning on image denoising: An overview.

Recently, deep convolutional neural networks have been applied in numerous image processing researches and have exhibited drastically improved performances. In this study, we introduce a densely connected hierarchical image denoising network (DHDN), which exceeds the performances of state-of-the-art image denoising solutions. Our proposed network improves the image denoising performance by applying the hierarchical architecture of the modified U-Net; this makes our network to use a larger number of parameters than other methods. In addition, we induce feature reuse and solve the vanishing-gradient problem by applying dense connectivity and residual learning to our convolution blocks and network. Finally, we successfully apply the model ensemble and self-ensemble methods; this enable us to improve the performance of the proposed network. The performance of the proposed network is validated by winning the second place in the NTRIE 2019 real image denoising challenge sRGB track and the third place in the raw-RGB track. Additional experimental results on additive white Gaussian noise removal also establishes that the proposed network outperforms conventional methods; this is notwithstanding the fact that the proposed network handles a wide range of noise levels with a single set of trained parameters.

/pdf/densely-connected-hierarchical-network-for-image-denoising-142yt9zvxo.pdf

Densely Connected Hierarchical Network for Image Denoising

A comprehensive survey on impulse and Gaussian denoising filters for digital images

Semantic edge detection (SED), which aims at jointly extracting edges as well as their category information, has far-reaching applications in domains such as semantic segmentation, object proposal generation, and object recognition. SED naturally requires achieving two distinct supervision targets: locating fine detailed edges and identifying high-level semantics. We shed light on how such distracted supervision targets prevent state-of-the-art SED methods from effectively using deep supervision to improve results. In this paper, we propose a novel fully convolutional neural network architecture using diverse deep supervision (DDS) within a multi-task framework where lower layers aim at generating category-agnostic edges, while higher layers are responsible for the detection of category-aware semantic edges. To overcome the distracted supervision challenge, a novel information converter unit is introduced, whose effectiveness has been extensively evaluated in several popular benchmark datasets, including SBD, Cityscapes, and PASCAL VOC2012. Source code will be released upon paper acceptance.

Semantic Edge Detection with Diverse Deep Supervision

Quality control is one of the industrial tasks most susceptible to be improved by implementing technological innovations. As an innovative technology, machine vision enables reliable and fast 24/7 inspections and helps producers to improve the efficiency of manufacturing operations. The accessible data by vision equipment will be used to identify and report defective products, understand the causes of deficiencies and allow rapid and efficient intervention in smart factories. From this perspective, the proposed machine vision model in this paper combines the identification of defective products and the continuous improvement of manufacturing processes by predicting the most suitable parameters of production processes to obtain a defect-free item. The suggested model exploits all generated data by various integrated technologies in the manufacturing chain, thus meeting the requirements of quality management in the context of Industry 4.0, based on predictive analysis to identify patterns in data and suggest corrective actions to ensure product quality. In addition, a comparative study between several machine learning algorithms, both for product classification and process improvement models, is performed in order to evaluate the designed system. The results of this study show that the proposed model largely meets the requirements for the proper implementation of these techniques.

https://www.mdpi.com/2224-2708/10/1/7/pdf

Intelligent Machine Vision Model for Defective Product Inspection Based on Machine Learning

This paper introduces a robust edge detection method that relies on an integrated process for denoising images in the presence of high impulse noise. This process is shown to be resilient to impulse (or salt and pepper) noise even under high intensity levels. The proposed switching adaptive median and fixed weighted mean filter (SAMFWMF) is shown to yield optimal edge detection and edge detail preservation, an outcome we validate through high correlation, structural similarity index, and peak signal-to-noise ratio measures. For comparative purposes, a comprehensive analysis of other denoising filters is provided based on these various validation metrics. The non-maximum suppression method and new edge following maximum-sequence are the two techniques used to track the edges and overcome edge discontinuities and noisy pixels, especially in the presence of high-intensity noise levels. After applying predefined thresholds to the grayscale image, and thus obtaining a binary image, several morphological operations are used to remove the unwanted edges and noisy pixels and perform edge thinning to ultimately provide the desired edge connectivity, which results in an optimal edge detection method. The obtained results are compared with other existing state-of-the-art denoising filters and other edge detection methods in support of our assertion that the proposed method is resilient to impulse noise even under high-intensity levels.

A Robust Edge Detection Approach in the Presence of High Impulse Noise Intensity Through Switching Adaptive Median and Fixed Weighted Mean Filtering

This study introduces an image denoising method in the presence of combined speckle and Gaussian noise The dual-tree complex wavelet transform is applied to the image in order to obtain specific coefficients characterising these types of noise Then, these extracted coefficients are removed by thresholding and an inverse wavelet transform is applied in order to obtain the reconstructed image A comparison between the dual-tree and standard wavelet-based denoising filters is provided on the basis of different structural metrics Finally, in order to remove any remaining noise, a spatial denoising filter is applied to the image The results obtained on medical ultrasound images corrupted by this noise combination support the authors' assertion on the method's resilience to the combined effects of speckle and Gaussian noise, and are compared to well-known and most effective speckle-Gaussian denoising filters

https://ietresearch.onlinelibrary.wiley.com/doi/epdf/10.1049/iet-ipr.2018.5292

Denoising of ultrasound images affected by combined speckle and Gaussian noise

In this paper, fatigue life of solder joints of the power module and the printed circuit board in a power inverter under directional random vibration using the finite element method (FEM) was evaluated. Five angles of 0°, 30°, 45°, 60°, and 90° were selected for direction of vibration loading. The simulation results showed that the outermost corner of the solder layer undergoes the maximum stress during the vibration frequencies in all the situations. In general, with the increase of input frequency, the solder joints got more susceptible to the catastrophic failure. Moreover, it was found that in lower angles of loading direction, the peeling stress is dramatically increased in the solder layer. This event can be attributed to the considerable portion of shear stress in loading directions with lower angles. The experimental results also proved the FEM simulation and showed that the void growth and coalescence, resulted by fatigue, are intensified in the lower angles of vibration loading. Overall, this paper indicates that the reliability of a power module under a normal loading (90°) is the highest among all other conditions.

Influence of Directional Random Vibration on the Fatigue Life of Solder Joints in a Power Module

This study introduces an image denoising method which focuses on detail preservation in the presence of high impulse (salt and pepper) noise. The proposed adaptive median and fixed weighted mean filter (AMFWMF) result in enhanced image similarity and optimal edge information preservation with high correlation and structural similarity index measures. For comparative purposes, a comprehensive analysis of other denoising filters is provided based on different structural metrics. Such standard measures, are used as standard measurements to gauge which of these methods leads to an optimal outcome. The provided results are compared to other existing denoising filters and support of hypothesis of a filter with high resilience to impulse noise of high density levels, our assertion on the method's resilience to impulse noise even under high-density levels. Then, we apply it to a set of MRI images corrupted by different levels of impulse noise intensity.

Mehdi Mafi

Papers

A comprehensive survey on impulse and Gaussian denoising filters for digital images

A Robust Edge Detection Approach in the Presence of High Impulse Noise Intensity Through Switching Adaptive Median and Fixed Weighted Mean Filtering

Denoising of ultrasound images affected by combined speckle and Gaussian noise

Influence of Directional Random Vibration on the Fatigue Life of Solder Joints in a Power Module

High impulse noise intensity removal in MRI images