Irene Cheng

Frincy Clement, Ji Yang, Irene Cheng

Deep Neural Networks have often been called the black box because of the complex, deep architecture and non-transparency presented by the inner layers. There is a lack of trust to use Artificial Intelligence in critical and high-precision fields such as security, finance, health, and manufacturing industries. A lot of focused work has been done to provide interpretable models, intending to deliver meaningful insights into the thoughts and behavior of neural networks. In our research, we compare the state-of-the-art methods in the Activation-based methods (ABM) for interpreting predictions of CNN models, specifically in the application of Image Classification. We then extend the same for eight CNN-based architectures to compare the differences in visualization and thus interpretability. We introduced a novel technique Feature CAM, which falls in the perturbation-activation combination, to create fine-grained, class-discriminative visualizations. The resulting saliency maps from our experiments proved to be 3-4 times better human interpretable than the state-of-the-art in ABM. At the same time it reserves machine interpretability, which is the average confidence scores in classification.

Frincy Clement, Kirtan Shah, Dhara Pancholi et al.

Textureless object recognition has become a significant task in Computer Vision with the advent of Robotics and its applications in manufacturing sector. It has been challenging to obtain good accuracy in real time because of its lack of discriminative features and reflectance properties which makes the techniques for textured object recognition insufficient for textureless objects. A lot of work has been done in the last 20 years, especially in the recent 5 years after the TLess and other textureless dataset were introduced. In this project, by applying image processing techniques we created a robust augmented dataset from initial imbalanced smaller dataset. We extracted edge features, feature combinations and RGB images enhanced with feature/feature combinations to create 15 datasets, each with a size of ~340,000. We then trained four classifiers on these 15 datasets to arrive at a conclusion as to which dataset performs the best overall and whether edge features are important for textureless objects. Based on our experiments and analysis, RGB images enhanced with combination of 3 edge features performed the best compared to all others. Model performance on dataset with HED edges performed comparatively better than other edge detectors like Canny or Prewitt.

Jinjiang Guo, Vincent Vidal, Irene Cheng et al.

Objective visual quality assessment of 3D models is a fundamental issue in computer graphics. Quality assessment metrics may allow a wide range of processes to be guided and evaluated, such as level of detail creation, compression, filtering, and so on. Most computer graphics assets are composed of geometric surfaces on which several texture images can be mapped to 11 make the rendering more realistic. While some quality assessment metrics exist for geometric surfaces, almost no research has been conducted on the evaluation of texture-mapped 3D models. In this context, we present a new subjective study to evaluate the perceptual quality of textured meshes, based on a paired comparison protocol. We introduce both texture and geometry distortions on a set of 5 reference models to produce a database of 136 distorted models, evaluated using two rendering protocols. Based on analysis of the results, we propose two new metrics for visual quality assessment of textured mesh, as optimized linear combinations of accurate geometry and texture quality measurements. These proposed perceptual metrics outperform their counterparts in terms of correlation with human opinion. The database, along with the associated subjective scores, will be made publicly available online.

Tahjid Ashfaque Mostafa, Irene Cheng

In this work, we explore various neural network architectures using Magnetic Resonance (MR) T1 images of the brain to identify Parkinson's Disease (PD), which is one of the most common neurodegenerative and movement disorders. We propose three ensemble architectures combining some winning Convolutional Neural Network models of ImageNet Large Scale Visual Recognition Challenge (ILSVRC). All of our proposed architectures outperform existing approaches to detect PD from MR images, achieving upto 95\% detection accuracy. We also find that when we construct our ensemble architecture using models pretrained on the ImageNet dataset unrelated to PD, the detection performance is significantly better compared to models without any prior training. Our finding suggests a promising direction when no or insufficient training data is available.

Tahjid Ashfaque Mostafa, Irene Cheng

Parkinson's Disease(PD) is one of the major nervous system disorders that affect people over 60. PD can cause cognitive impairments. In this work, we explore various approaches to identify Parkinson's using Magnetic Resonance (MR) T1 images of the brain. We experiment with ensemble architectures combining some winning Convolutional Neural Network models of ImageNet Large Scale Visual Recognition Challenge (ILSVRC) and propose two architectures. We find that detection accuracy increases drastically when we focus on the Gray Matter (GM) and White Matter (WM) regions from the MR images instead of using whole MR images. We achieved an average accuracy of 94.7\% using smoothed GM and WM extracts and one of our proposed architectures. We also perform occlusion analysis and determine which brain areas are relevant in the architecture decision making process.

Subhayan Mukherjee, Aaron Zimmer, Xinyao Sun et al.

Phase filtering and pixel quality (coherence) estimation is critical in producing Digital Elevation Models (DEMs) from Interferometric Synthetic Aperture Radar (InSAR) images, as it removes spatial inconsistencies (residues) and immensely improves the subsequent unwrapping. Large amount of InSAR data facilitates Wide Area Monitoring (WAM) over geographical regions. Advances in parallel computing have accelerated Convolutional Neural Networks (CNNs), giving them advantages over human performance on visual pattern recognition, which makes CNNs a good choice for WAM. Nevertheless, this research is largely unexplored. We thus propose "GenInSAR", a CNN-based generative model for joint phase filtering and coherence estimation, that directly learns the InSAR data distribution. GenInSAR's unsupervised training on satellite and simulated noisy InSAR images outperforms other five related methods in total residue reduction (over 16.5% better on average) with less over-smoothing/artefacts around branch cuts. GenInSAR's Phase, and Coherence Root-Mean-Squared-Error and Phase Cosine Error have average improvements of 0.54, 0.07, and 0.05 respectively compared to the related methods.

Subhayan Mukherjee, Guan-Ming Su, Irene Cheng

High Dynamic Range (HDR) imaging is gaining increased attention due to its realistic content, for not only regular displays but also smartphones. Before sufficient HDR content is distributed, HDR visualization still relies mostly on converting Standard Dynamic Range (SDR) content. SDR images are often quantized, or bit depth reduced, before SDR-to-HDR conversion, e.g. for video transmission. Quantization can easily lead to banding artefacts. In some computing and/or memory I/O limited environment, the traditional solution using spatial neighborhood information is not feasible. Our method includes noise generation (offline) and noise injection (online), and operates on pixels of the quantized image. We vary the magnitude and structure of the noise pattern adaptively based on the luma of the quantized pixel and the slope of the inverse-tone mapping function. Subjective user evaluations confirm the superior performance of our technique.

Subhayan Mukherjee, Navaneeth Kamballur Kottayil, Xinyao Sun et al.

We propose a novel direction to improve the denoising quality of filtering-based denoising algorithms in real time by predicting the best filter parameter value using a Convolutional Neural Network (CNN). We take the use case of BM3D, the state-of-the-art filtering-based denoising algorithm, to demonstrate and validate our approach. We propose and train a simple, shallow CNN to predict in real time, the optimum filter parameter value, given the input noisy image. Each training example consists of a noisy input image (training data) and the filter parameter value that produces the best output (training label). Both qualitative and quantitative results using the widely used PSNR and SSIM metrics on the popular BSD68 dataset show that the CNN-guided BM3D outperforms the original, unguided BM3D across different noise levels. Thus, our proposed method is a CNN-based improvement on the original BM3D which uses a fixed, default parameter value for all images.

Subhayan Mukherjee, Aaron Zimmer, Xinyao Sun et al.

Interferometric Synthetic Aperture Radar (InSAR) imagery based on microwaves reflected off ground targets is becoming increasingly important in remote sensing for ground movement estimation. However, the reflections are contaminated by noise, which distorts the signal's wrapped phase. Demarcation of image regions based on degree of contamination ("coherence") is an important component of the InSAR processing pipeline. We introduce Convolutional Neural Networks (CNNs) to this problem domain and show their effectiveness in improving coherence-based demarcation and reducing misclassifications in completely incoherent regions through intelligent preprocessing of training data. Quantitative and qualitative comparisons prove superiority of proposed method over three established methods.

Subhayan Mukherjee, Aaron Zimmer, Navaneeth Kamballur Kottayil et al.

Interferometric Synthetic Aperture Radar (InSAR) imagery for estimating ground movement, based on microwaves reflected off ground targets is gaining increasing importance in remote sensing. However, noise corrupts microwave reflections received at satellite and contaminates the signal's wrapped phase. We introduce Convolutional Neural Networks (CNNs) to this problem domain and show the effectiveness of autoencoder CNN architectures to learn InSAR image denoising filters in the absence of clean ground truth images, and for artefact reduction in estimated coherence through intelligent preprocessing of training data. We compare our results with four established methods to illustrate superiority of proposed method.

Subhayan Mukherjee, Giuseppe Valenzise, Irene Cheng

Image Quality Assessment algorithms predict a quality score for a pristine or distorted input image, such that it correlates with human opinion. Traditional methods required a non-distorted "reference" version of the input image to compare with, in order to predict this score. However, recent "No-reference" methods circumvent this requirement by modelling the distribution of clean image features, thereby making them more suitable for practical use. However, majority of such methods either use hand-crafted features or require training on human opinion scores (supervised learning), which are difficult to obtain and standardise. We explore the possibility of using deep features instead, particularly, the encoded (bottleneck) feature maps of a Convolutional Autoencoder neural network architecture. Also, we do not train the network on subjective scores (unsupervised learning). The primary requirements for an IQA method are monotonic increase in predicted scores with increasing degree of input image distortion, and consistent ranking of images with the same distortion type and content, but different distortion levels. Quantitative experiments using the Pearson, Kendall and Spearman correlation scores on a diverse set of images show that our proposed method meets the above requirements better than the state-of-art method (which uses hand-crafted features) for three types of distortions: blurring, noise and compression artefacts. This demonstrates the potential for future research in this relatively unexplored sub-area within IQA.

Xinyao Sun, Aaron Zimmer, Subhayan Mukherjee et al.

Over the past decade, Interferometric Synthetic Aperture Radar (InSAR) has become a successful remote sensing technique. However, during the acquisition step, microwave reflections received at satellite are usually disturbed by strong noise, leading to a noisy single-look complex (SLC) SAR image. The quality of their interferometric phase is even worse. InSAR phase filtering is an ill-posed problem and plays a key role in subsequent processing. However, most of existing methods usually require expert supervision or heavy runtime, which limits the usability and scalability for practical usages such as wide-area monitoring and forecasting. In this work, we propose a deep convolutional neural network (CNN) based model DeepInSAR to intelligently solve both the phase filtering and coherence estimation problems. We demonstrate our DeepInSAR using both simulated and real data. A teacher-student framework is proposed to deal with the issue that there is no ground truth sample for real-world InSAR data. Quantitative and qualitative comparisons show that DeepInSAR achieves comparable or even better results than its stacked-based teacher method on new test datasets but requiring fewer pairs of SLCs as well as outperforms three other established non-stack based methods with less running time and no human supervision.

Tao Wang, Irene Cheng, Anup Basu

Brain tumor segmentation from Magnetic Resonance Images (MRIs) is an important task to measure tumor responses to treatments. However, automatic segmentation is very challenging. This paper presents an automatic brain tumor segmentation method based on a Normalized Gaussian Bayesian classification and a new 3D Fluid Vector Flow (FVF) algorithm. In our method, a Normalized Gaussian Mixture Model (NGMM) is proposed and used to model the healthy brain tissues. Gaussian Bayesian Classifier is exploited to acquire a Gaussian Bayesian Brain Map (GBBM) from the test brain MR images. GBBM is further processed to initialize the 3D FVF algorithm, which segments the brain tumor. This algorithm has two major contributions. First, we present a NGMM to model healthy brains. Second, we extend our 2D FVF algorithm to 3D space and use it for brain tumor segmentation. The proposed method is validated on a publicly available dataset.

Subhayan Mukherjee, Irene Cheng, Steven Miller et al.

White Matter Injury (WMI) is the most prevalent brain injury in the preterm neonate leading to developmental deficits. However, detecting WMI in Magnetic Resonance (MR) images of preterm neonate brains using traditional WM segmentation-based methods is difficult mainly due to lack of reliable preterm neonate brain atlases to guide segmentation. Hence, we propose a segmentation-free, fast, unsupervised, atlas-free WMI detection method. We detect the ventricles as blobs using a fast linear Maximally Stable Extremal Regions algorithm. A reference contour equidistant from the blobs and the brain-background boundary is used to identify tissue adjacent to the blobs. Assuming normal distribution of the gray-value intensity of this tissue, the outlier intensities in the entire brain region are identified as potential WMI candidates. Thereafter, false positives are discriminated using appropriate heuristics. Experiments using an expert-annotated dataset show that the proposed method runs 20 times faster than our earlier work which relied on time-consuming segmentation of the WM region, without compromising WMI detection accuracy.

Sara Soltaninejad, Irene Cheng, Anup Basu

Parkinson's Disease (PD) is one of the most common types of neurological diseases caused by progressive degeneration of dopamin- ergic neurons in the brain. Even though there is no fixed cure for this neurodegenerative disease, earlier diagnosis followed by earlier treatment can help patients have a better quality of life. Magnetic Resonance Imag- ing (MRI) has been one of the most popular diagnostic tool in recent years because it avoids harmful radiations. In this paper, we investi- gate the plausibility of using MRIs for automatically diagnosing PD. Our proposed method has three main steps : 1) Preprocessing, 2) Fea- ture Extraction, and 3) Classification. The FreeSurfer library is used for the first and the second steps. For classification, three main types of classifiers, including Logistic Regression (LR), Random Forest (RF) and Support Vector Machine (SVM), are applied and their classification abil- ity is compared. The Parkinsons Progression Markers Initiative (PPMI) data set is used to evaluate the proposed method. The proposed system prove to be promising in assisting the diagnosis of PD.

Mehdi Faraji, Irene Cheng, Iris Naudin et al.

Intravascular Ultrasound (IVUS) is an intra-operative imaging modality that facilitates observing and appraising the vessel wall structure of the human coronary arteries. Segmentation of arterial wall boundaries from the IVUS images is not only crucial for quantitative analysis of the vessel walls and plaque characteristics, but is also necessary for generating 3D reconstructed models of the artery. The aim of this study is twofold. Firstly, we investigate the feasibility of using a recently proposed region detector, namely Extremal Region of Extremum Level (EREL) to delineate the luminal and media-adventitia borders in IVUS frames acquired by 20 MHz probes. Secondly, we propose a region selection strategy to label two ERELs as lumen and media based on the stability of their textural information. We extensively evaluated our selection strategy on the test set of a standard publicly available dataset containing 326 IVUS B-mode images. We showed that in the best case, the average Hausdorff Distances (HD) between the extracted ERELs and the actual lumen and media were $0.22$ mm and $0.45$ mm, respectively. The results of our experiments revealed that our selection strategy was able to segment the lumen with $\le 0.3$ mm HD to the gold standard even though the images contained major artifacts such as bifurcations, shadows, and side branches. Moreover, when there was no artifact, our proposed method was able to delineate media-adventitia boundaries with $0.31$ mm HD to the gold standard. Furthermore, our proposed segmentation method runs in time that is linear in the number of pixels in each frame. Based on the results of this work, by using a 20 MHz IVUS probe with controlled pullback, not only can we now analyze the internal structure of human arteries more accurately, but also segment each frame during the pullback procedure because of the low run time of our proposed segmentation method.

Navaneeth Kamballur Kottayil, Giuseppe Valenzise, Frederic Dufaux et al.

Accurate prediction of local distortion visibility thresholds is critical in many image and video processing applications. Existing methods require an accurate modeling of the human visual system, and are derived through pshycophysical experiments with simple, artificial stimuli. These approaches, however, are difficult to generalize to natural images with complex types of distortion. In this paper, we explore a different perspective, and we investigate whether it is possible to learn local distortion visibility from image quality scores. We propose a convolutional neural network based optimization framework to infer local detection thresholds in a distorted image. Our model is trained on multiple quality datasets, and the results are correlated with empirical visibility thresholds collected on complex stimuli in a recent study. Our results are comparable to state-of-the-art mathematical models that were trained on phsycovisual data directly. This suggests that it is possible to predict psychophysical phenomena from visibility information embedded in image quality scores.

Navaneeth Kamballur Kottayil, Giuseppe Valenzise, Frederic Dufaux et al.

Assessing the visual quality of High Dynamic Range (HDR) images is an unexplored and an interesting research topic that has become relevant with the current boom in HDR technology. We propose a new convolutional neural network based model for No reference image quality assessment(NR-IQA) on HDR data. This model predicts the amount and location of noise, perceptual influence of image pixels on the noise, and the perceived quality, of a distorted image without any reference image. The proposed model extracts numerical values corresponding to the noise present in any given distorted image, and the perceptual effects exhibited by a human eye when presented with the same. These two measures are extracted separately yet sequentially and combined in a mixing function to compute the quality of the distorted image perceived by a human eye. Our training process derives the the component that computes perceptual effects from a real world image quality dataset, rather than using results of psycovisual experiments. With the proposed model, we demonstrate state of the art performance for HDR NR-IQA and our results show performance similar to HDR Full Reference Image Quality Assessment algorithms (FR-IQA).

Navaneeth Kamballur Kottayil, Rositsa Bogdanova, Irene Cheng et al.

Laparoscopic Surgery (LS) is a modern surgical technique whereby the surgery is performed through an incision with tools and camera as opposed to conventional open surgery. This promises minimal recovery times and less hemorrhaging. Multi view LS is the latest development in the field, where the system uses multiple cameras to give the surgeon more information about the surgical site, potentially making the surgery easier. In this publication, we study the gaze patterns of a high performing subject in a multi-view LS environment and compare it with that of a novice to detect the differences between the gaze behavior. This was done by conducting a user study with 20 university students with varying levels of expertise in Multi-view LS. The subjects performed an laparoscopic task in simulation with three cameras (front/top/side). The subjects were then separated as high and low performers depending on the performance times and their data was analyzed. Our results show statistically significant differences between the two behaviors. This opens up new areas from of training novices to Multi-view LS to making smart displays that guide your shows the optimum view depending on the situation.

Navaneeth K. Kottayil, Irene Cheng, Frederic Dufaux et al.

Image Quality Assessment (IQA) algorithms evaluate the perceptual quality of an image using evaluation scores that assess the similarity or difference between two images. We propose a new low-level feature based IQA technique, which applies filter-bank decomposition and center-surround methodology. Differing from existing methods, our model incorporates color intensity adaptation and frequency scaling optimization at each filter-bank level and spatial orientation to extract and enhance perceptually significant features. Our computational model exploits the concept of object detection and encapsulates characteristics proposed in other IQA algorithms in a unified architecture. We also propose a systematic approach to review the evolution of IQA algorithms using unbiased test datasets, instead of looking at individual scores in isolation. Experimental results demonstrate the feasibility of our approach.

Subhayan Mukherjee, Irene Cheng, Anup Basu

Stereo images have been captured primarily for 3D reconstruction in the past. However, the depth information acquired from stereo can also be used along with saliency to highlight certain objects in a scene. This approach can be used to make still images more interesting to look at, and highlight objects of interest in the scene. We introduce this novel direction in this paper, and discuss the theoretical framework behind the approach. Even though we use depth from stereo in this work, our approach is applicable to depth data acquired from any sensor modality. Experimental results on both indoor and outdoor scenes demonstrate the benefits of our algorithm.

Subhayan Mukherjee, Irene Cheng, Ram Mohana Reddy Guddeti et al.

Stereo depth estimation is error-prone; hence, effective error detection methods are desirable. Most such existing methods depend on characteristics of the stereo matching cost curve, making them unduly dependent on functional details of the matching algorithm. As a remedy, we propose a novel error detection approach based solely on the input image and its depth map. Our assumption is that, entropy of any point on an image will be significantly higher than the entropy of its corresponding point on the image's depth map. In this paper, we propose a confidence measure, Entropy-Difference (ED) for stereo depth estimates and a binary classification method to identify incorrect depths. Experiments on the Middlebury dataset show the effectiveness of our method. Our proposed stereo confidence measure outperforms 17 existing measures in all aspects except occlusion detection. Established metrics such as precision, accuracy, recall, and area-under-curve are used to demonstrate the effectiveness of our method.