Unsupervised Image-Based Classification of Corrosion Severity in Automobile Engine Connecting Rods

The objective of this paper is to develop an efficient noise detector and filter for random valued impulse noise (RVIN) in gray level images. A Dual Histogram based Noise Detector (DHND) is proposed and compared with other approaches using the classical measures like False Alarm (FA) and Miss Detection (MD). Also a novel filtering approach is proposed namely Hybrid Gaussian Filtering (HGF) which combines two filters such as Gaussian Impulse Kernel (GIK) and Gaussian Weighted Median (GWM). The proposed HGF approach is validated across state-of-art filters in terms of standard performance measures such as Peak Signal to Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM) and Feature Similarity Index Measure (FSIM). The robustness of the filter is experimented for various standard gray scale images to evaluate its global significance. Experimental results reveal the pros and cons of other algorithms along with the effectiveness of the proposed filter under varying noise conditions.

Thin clouds seriously affect the availability of optical remote sensing images, especially in visible bands. Short-wave infrared (SWIR) bands are less influenced by thin clouds, but usually have lower spatial resolution than visible (Vis) bands in high spatial resolution remote sensing images (e.g., in Sentinel-2A/B, CBERS04, ZY-1 02D and HJ-1B satellites). Most cloud removal methods do not take advantage of the spectral information available in SWIR bands, which are less affected by clouds, to restore the background information tainted by thin clouds in Vis bands. In this paper, we propose CR-MSS, a novel deep learning-based thin cloud removal method that takes the SWIR and vegetation red edge (VRE) bands as inputs in addition to visible/near infrared (Vis/NIR) bands, in order to improve cloud removal in Sentinel-2 visible bands. Contrary to some traditional and deep learning-based cloud removal methods, which use manually designed rescaling algorithm to handle bands at different resolutions, CR-MSS uses convolutional layers to automatically process bands at different resolution. CR-MSS has two input/output branches that are designed to process Vis/NIR and VRE/SWIR, respectively. Firstly, Vis/NIR cloudy bands are down-sampled by a convolutional layer to low spatial resolution features, which are then concatenated with the corresponding features extracted from VRE/SWIR bands. Secondly, the concatenated features are put into a fusion tunnel to down-sample and fuse the spectral information from Vis/NIR and VRE/SWIR bands. Third, a decomposition tunnel is designed to up-sample and decompose the fused features. Finally, a transpose convolutional layer is used to up-sample the feature maps to the resolution of input Vis/NIR bands. CR-MSS was trained on 28 real Sentinel-2A image pairs over the globe, and tested separately on eight real cloud image pairs and eight simulated cloud image pairs. The average SSIM values (Structural Similarity Index Measurement) for CR-MSS results on Vis/NIR bands over all testing images were 0.69, 0.71, 0.77, and 0.81, respectively, which was on average 1.74% higher than the best baseline method. The visual results on real Sentinel-2 images demonstrate that CR-MSS can produce more realistic cloud and cloud shadow removal results than baseline methods.

Cone-beam computed tomography (CBCT) scans, performed fractionally (e.g., daily or weekly), are widely utilized for patient alignment in the image-guided radiotherapy (IGRT) process, thereby making it a potential imaging modality for the implementation of adaptive radiotherapy (ART) protocols. Nonetheless, significant artifacts and incorrect Hounsfield unit (HU) values hinder their application in quantitative tasks such as target and organ segmentations and dose calculation. Therefore, acquiring CT-quality images from the CBCT scans is essential to implement online ART in clinical settings. This work aims to develop an unsupervised learning method using the patient-specific diffusion model for CBCT-based synthetic CT (sCT) generation to improve the image quality of CBCT. The proposed method is in an unsupervised framework that utilizes a patient-specific score-based model as the image prior alongside a customized total variation (TV) regularization to enforce coherence across different transverse slices. The score-based model is unconditionally trained using the same patient's planning CT (pCT) images to characterize the manifold of CT-quality images and capture the unique anatomical information of the specific patient. The efficacy of the proposed method was assessed on images from anatomical sites including head and neck (H&N) cancer, pancreatic cancer, and lung cancer. The performance of the proposed CBCT correction method was evaluated using quantitative metrics, including mean absolute error (MAE), non-uniformity (NU), and structural similarity index measure (SSIM). Additionally, the proposed algorithm was benchmarked against other unsupervised learning-based CBCT correction algorithms. The proposed method significantly reduced various kinds of CBCT artifacts in the studies of H&N, pancreatic, and lung cancer patients. In the lung stereotactic body radiation therapy (SBRT) patient study, the MAE, NU, and SSIM were improved from 47 HU, 45 HU, and 0.58 in the original CBCT images to 13 HU, 14dB, and 0.67 in the generated sCT images. Compared to other unsupervised learning-based algorithms, the proposed method demonstrated superior performance in artifact reduction. The proposed unsupervised method can generate sCT from CBCT with reduced artifacts and precise HU values, enabling CBCT-guided segmentation and replanning for online ART.

Facial recognition is a very hot topic due to its importance in security, surveillance systems, law enforcement, etc. Many methods are proposed to achieve high accuracy in face recognition but each method has its weaknesses as well as its points of strength. In this paper, we present an effective similarity measure to recognize human faces this measure is called (KSM). The proposed measure is based on the statistical properties of the image by using the high-order statistics (HOS), specifically kurtosis and skewness, for facial recognition. Performance evaluation was conducted using face images from AT&T database and FEI (Brazilian) database. The simulation results show that the proposed measure KSM outperforms the well-known Structural Similarity Index Measure (SSIM) and Feature Similarity Index Measure (FSIM) by the ability to detect similarity even under the difference in illumination, facial expression and pose variations.

To implement image-guided adaptive radiotherapy (IGART), many studies investigated dose calculations on cone-beam computed tomography (CBCT). A high HU accuracy is crucial for a high dose calculation accuracy and many imaging sites showed satisfactory results. It has been shown that the dose calculation accuracy for lung cancer lags behind. To examine why the dose calculation accuracy for lung is insufficient, the relative effects of the field-of-view (FOV), breathing motion, and scatter on dose calculation accuracy were studied. A framework was built to simulate CBCT scans for lung cancer patients by forward projecting repeat CT (rCT) scans for two scan geometries: small (SFOV) and medium FOV (MFOV). Breathing motion was modeled by applying a 4D deformation vector field to the mid-position rCT. Scatter was modeled by Monte-Carlo simulations with/without an anti-scatter grid (ASG). Simulated projections were reconstructed using filtered back-projection with/without scatter correction. In case of the SFOV, the CBCT images were patched with the planning CT scan in axial direction. The treatment plan was recalculated on the rCT and simulated CBCT. The mean Hounsfield unit (HU) difference (ΔHUmean), the structural similarity index measure (SSIM), and γ metrics were calculated for the CBCT datasets of various imaging settings. The differences in HU, SSIM and dose calculation accuracy for CBCTs with and without breathing motion were negligible (mean ΔHUmean=6.4vs. 13.7, mean SSIM=0.941vs. 0.957, mean γ (ref=MFOV)=0.75). The SFOV resulted in a lower HU (mean ΔHUmean=-9.2vs. 13.7) and SSIM (mean SSIM=0.912vs. 0.957), and therefore in dose differences compared to the MFOV (mean γ=1.22). Scatter led to considerable discrepancies in all metrics. Adding only the ASG improved the results more than only applying a scatter correction algorithm. Combining ASG and scatter correction algorithm resulted in an even higher dose calculation accuracy. Scatter and FOV are the main contributors to dose inaccuracies and motion has only a minor effect on dose calculation accuracy. Therefore, utilizing an appropriate scatter correction and FOV is important to achieve sufficient dose calculation accuracy to facilitate IGART for lung.

Image segmentation using bi-level thresholds works well for straightforward scenarios; however, dealing with complex images that contain multiple objects or colors presents considerable computational difficulties. Multi-level thresholding is crucial for these situations, but it also introduces a challenging optimization problem. This paper presents an improved Reptile Search Algorithm (RSA) that includes a Gbest operator to enhance its performance. The proposed method determines optimal threshold values for both grayscale and color images, utilizing entropy-based objective functions derived from the Otsu and Kapur techniques. Experiments were carried out on 16 benchmark images, which included COVID-19 scans along with standard color and grayscale images. A thorough evaluation was conducted using metrics such as the fitness function, peak signal-to-noise ratio (PSNR), structural similarity index measure (SSIM), and the Friedman ranking test. The results indicate that the proposed algorithm seems to surpass existing state-of-the-art methods, demonstrating its effectiveness and robustness in multi-level thresholding tasks.

Image denoising has become one of the fundamental problems in the field of image processing. We present an iterative image denoising method named as multitexton noise identification and local dissimilarity-based noise removal that incorporates a noise restoration phase followed by a noise identification phase for gray-scale images corrupted by random-valued impulse noise (RVIN). Multiple textons of distinct orientations arranged on the basis of radial symmetry are proposed that give sharp edges of restored images. The noise identification phase works on an adaptive threshold range by employing the local statistics of textons. Moreover, the proposed method can better be utilized to deal with the uncertainty observed in identification of noisy and edge pixels. The dissimilarity of the identified corrupted pixel with its four-connected neighboring pixels is computed by considering the similarity among these four-connected pixels, to restore the gray-level intensities of identified corrupted pixels at the second phase. Standard gray-scale benchmark test images are used to assess the quantitative and visual performance of the proposed method by comparing it with state-of-the-art denoising methods. The extensive experimental results reveal that the presented method outperforms in terms of peak-signal-to-noise ratio, structural similarity index measurement, miss detection, false detection, and visual results for both lower and higher intensities of RVIN as compared to other existing techniques.

A New measure is proposed for assessing the similarity among gray-scale images. The well-known Structural Similarity Index Measure (SSIM) has been designed using a statistical approach that fails under significant noise (lowPSNR). The proposed measure, denoted by Manhattan distance and STD, uses a combination of two parts: the first part is the Geometric method, while the second part is based on the statistical feature. The concept of manhattan distance is used in the geometric part. The new measure shows the advantages of statistical approaches and geometric approaches. The proposed similarity method is an outcome for the human face. The novel measure outperforms the classical SSIM in detecting image similarity at low PSNR, with a significant difference in performance. AMS subject classification:

ABSTRACTIn this study, the combination of surface reflectance products from Terra- Moderate Resolution Imaging Spectroradiometer and Landsat-Enhanced Thematic Mapper Plus sensors are explored through the Flexible Spatiotemporal DAta Fusion (FSDAF) algorithm within the framework of forest fire studies over tropical savannah environments. Thus, 60 fusion-derived images were generated from four spectral bands [red, near-infrared, shortwave infrared (SWIR1 and SWIR2)] and six spectral indices [normalized difference vegetation index, normalized difference moisture index, global environment monitoring index, soil-adjusted vegetation index, normalized burn ratio (NBR), and differenced normalized burn ratio (dNBR)] over two selected study sites. For all fusion processes performed, the actual Landsat images for the corresponding dates are available, which supports validation of the blended images. Additionally, integration of blended spectral indices in the immediate post-fire evaluation and the generation of fire severity were analysed. The blended bands presented correlation and Structure Similarity Index Measure (SSIM) values that were consistently higher than 0.819 and root mean square error values of less than 0.027, which confirms good accuracy levels obtained from the model. Similar correlation and SSIM accuracy levels were observed in the blended indices assessment for both study sites, which enables its values to be well-integrated for an analysis of the immediately post-fire date. However, the fire severity mapping from fused images needs to be carefully implemented since the dNBR index is generally less accurate than other blended indices. FSDAF fusion proved to be a useful alternative to retrieving multispectral information from savannah environments affected by fires.

In the field of remote sensing satellite imagery, malfunctions in the available raw data are prominent. Especially in Short-Wave Infrared (SWIR) detectors used in satellite imaging cameras, which suffer from dropouts in pixel and line direction in raw data. With the recent development in generative adversarial networks and its vast application in inpainting the missing data, the possibility to predict and fill in the missing data accurately with contextual attention has become prevalent. This paper presents SN-GANs (SN-generative adversarial networks) which is two-staged architecture, and it is based on the concept of feed forward neural networks with contextual attention layers. While reconstructing the corrupted part of the images, the model takes surrounding pixels into consideration. Moreover, this architecture is adept enough to fill in the multiple lines and pixel dropouts efficiently even in super-resolution satellite images. The available traditional methods fail to address the loss of data that incurs, while inpainting a 16-bit raw image because they are effective enough for 8-bit RGB images. SN-GANs have effectively resolved this issue with a lossless image inpainting method for 16-bit satellite images as it retains the features of non-corrupted data. The performance of the model is evaluated using similarity metrics like structural similarity index measure (SSIM), peak signal-to-noise ratio (PSNR) and mean-squared error (MSE).

Focus on test, measurement, quantum metrology, and analytical equipment

Ultrasonic testing has been widely used for nondestructive applications because it is inexpensive and has high inspection accuracy. However, massive ultrasonic data generated from the signal acquisition platform limits the performance of ultrasonic applications. Therefore, ultrasonic data compression with high accuracy is significant. In this study, we propose to use unsupervised learning (UL) to compress 3D ultrasonic data. The UL learns principal latent patterns and removes redundant information in ultrasonic data. The latent patterns are used to compress the 3D ultrasonic data. In this study, we propose to use several UL methods: Principal Component Analysis (PCA), Incremental-PCA, Independent Component Analysis (ICA), and Exploratory Factor Analysis (EFA) to learn latent patterns in acquired 3D ultrasonic data. An automatic ultrasonic testbed platform is assembled for ultrasonic data acquisition. By using UL, for high compression performance, we obtained the average 4.25% compression ratio; for high reconstruction performance, we obtained the average 9.67% compression ratio. The reconstruction accuracy is measured using the Peak Signal-to-Noise Ratio (PSNR) and Structural Similarity Index Measure (SSIM).

In the field of image processing, removal of noise from Gray scale as well as RGB images is an ambitious task. The important function of noise removal algorithm is to eliminate noise from a noisy image. The salt and pepper noise (SPN) is frequently arising into Gray scale and RGB images while capturing, acquiring and transmitting over the insecure several communication mechanisms. In past, the numerous noise removal methods have been introduced to extract the noise from images adulterated with SPN. The proposed work introduces the SPN removal algorithm for Gray scale at low along with high density noise (10\% to 90\%). According to the different conditions of proposed algorithm, the noisy pixel is reconstructed by Winsorized mean or mean value of all pixels except the centre pixel which are present in the processing window. The noise from an image can be removed by using the proposed algorithm without degrading the quality of image. The performance evaluation of proposed and modified decision based unsymmetric median filter (MDBUTMF) is done on the basis of different performance parameters such as Peak Signal to Noise Ratio (PSNR), Mean Square Error (MSE), Image Enhancement Factor (IEF) and Structure Similarity Index Measurement (SSIM).

Metastatic cancer is the leading cause of almost 90% of cancer-related deaths. Early detection of metastatic lesions in breast cancer would lead to judicious therapeutic choices in the management of the disease. Current clinical imaging techniques are unable to detect micro-metastases due to limited resolution. Optical imaging technologies are challenged by high tissue scattering and absorption of visible light, limiting resolution of deeper tissue lesions. While optical imaging modalities have the potential for real-time in vivo monitoring, poor penetration through biological mediums limits their application to subcutaneous lesions. Our approach utilizes rare earth (RE) nanoprobes that absorb near infrared (NIR) radiation and emit in the shortwave infrared (SWIR) spectrum (1000-3000 nm), allowing for greater depth of detection. In this study, we demonstrate the ability of CXCR4-targeting Rare Earth Albumin NanoComposites (ReANCs) to preferentially accumulate in receptor positive tumor lesions and image lesion dynamics in an in vivo MDA-MB-231 derived lung metastatic model. ReANCs or AMD3100 (small molecule inhibitor of CXCR4) functionalized ReANCs (fReANCs) were injected via the tail vein once a week and SWIR imaging was performed up to 48 hours post injection. Tumor burden in the lungs was assessed by MRI and compared to the SWIR signal from the nanoprobes. Longitudinal in vivo imaging confirmed improved accumulation of functionalized nanoprobes in tumor lesions compared to control ReANCs. This improved accumulation at early lesion stages allowed for earlier and more sensitive detection of internal tumors. We also observed control ReANCs cleared faster through the liver as evidenced by increased SWIR signal in the liver of these animals. In addition to the targeted imaging effect of fReANCs, we were able to demonstrate the ability of ReANCs to image micro-metastatic lesions in long bones and spine of animals prior to their detection via conventional imaging techniques such as MRI and CT. This was confirmed by ex-vivo SWIR imaging of the skeleton and micro-CT of the long bones. This advancement is exciting due to the paucity of imaging modalities capable of detecting micro-metastasis of breast cancer to bone. Histological analysis was performed to validate the presence of micro-lesions in the bones. Furthermore, immunohistochemical analysis of bone sections using markers over-expressed by MDA-MB-231 cells, namely Vimentin and pan-Cytokeratin confirmed the presence of tumors in the regions of interest.. Findings from this study support the promise of the “new window” imaging platform and suggest future clinical translatability for nanomedicine. Given the multifunctional nature of ANCs as drug delivery vehicles, our imaging probes can also be adapted for pre-clinical pharmacogenomic screening, an emerging component of precision medicine. Citation Format: Margot Zevon, Vidya Ganapathy, Harini Kantamneni, Marco Mingozzi, Paul Kim, Derek Adler, Mark Pierce, Richard Riman, Charles Roth, Prabhas Moghe. Early detection and longitudinal imaging of breast cancer metastatic microlesions using short-wave infrared light emitting rare-earth nanoprobes. [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr 1482. doi:10.1158/1538-7445.AM2015-1482

Optical metasurfaces (flat optics) allow unprecedented control over light, enabling multi‐dimensional light modulation. We propose a non‐local metasurface hosted by phase change material Sb2S3 for tunable image processing. It supports three imaging modalities: bright field, edge detection, and image denoising for intensity noise, functioning as diffractive image denoisers. The Structural Similarity Index Measure is used as a metric between the input noisy image and the denoised image. By tuning the phase of Sb2S3, its refractive index changes, effectively shifting the electromagnetic modes and resulting in these imaging modalities by providing the required optical transfer function (OTF). We optimized the metasurface design to achieve the required OTF and performed simulations on complex images with many corners and 2‐dimensional structures. We introduced salt and pepper noise into the input image and conducted simulations to evaluate performance. We discuss the shape of the OTF for image denoising applications and its adaptation for simultaneous image denoising and edge detection, both of which involve high spatial frequencies of the object. Our dynamically tunable metasurface platform can seamlessly integrate with standard coherent imaging systems, enabling versatile operations on the input image.