Search

One of the challenges in the study of Generative Adversarial Networks (GANs) is the difficulty of its performance control. Lipschitz constraint is essential in guaranteeing training stability for GANs. Although heuristic methods such as weight clipping, gradient penalty and spectral normalization have been proposed to enforce Lipschitz constraint, it is still difficult to achieve a solution that is both practically effective and theoretically provably satisfying a Lipschitz constraint. In this paper, we introduce the boundedness and continuity ($BC$) conditions to enforce the Lipschitz constraint on the discriminator functions of GANs. We prove theoretically that GANs with discriminators meeting the BC conditions satisfy the Lipschitz constraint. We present a practically very effective implementation of a GAN based on a convolutional neural network (CNN) by forcing the CNN to satisfy the $BC$ conditions (BC-GAN). We show that as compared to recent techniques including gradient penalty and spectral normalization, BC-GANs not only have better performances but also lower computational complexity., Comment: 20 pages, 7 figures

Motion blur, out of focus, insufficient spatial resolution, lossy compression and many other factors can all cause an image to have poor quality. However, image quality is a largely ignored issue in traditional pattern recognition literature. In this paper, we use face detection and recognition as case studies to show that image quality is an essential factor which will affect the performances of traditional algorithms. We demonstrated that it is not the image quality itself that is the most important, but rather the quality of the images in the training set should have similar quality as those in the testing set. To handle real-world application scenarios where images with different kinds and severities of degradation can be presented to the system, we have developed a quality classified image analysis framework to deal with images of mixed qualities adaptively. We use deep neural networks first to classify images based on their quality classes and then design a separate face detector and recognizer for images in each quality class. We will present experimental results to show that our quality classified framework can accurately classify images based on the type and severity of image degradations and can significantly boost the performances of state-of-the-art face detector and recognizer in dealing with image datasets containing mixed quality images., Comment: 6 pages

One of the methods for stratifying different molecular classes of breast cancer is the Nottingham Prognostic Index Plus (NPI+) which uses breast cancer relevant biomarkers to stain tumour tissues prepared on tissue microarray (TMA). To determine the molecular class of the tumour, pathologists will have to manually mark the nuclei activity biomarkers through a microscope and use a semi-quantitative assessment method to assign a histochemical score (H-Score) to each TMA core. Manually marking positively stained nuclei is a time consuming, imprecise and subjective process which will lead to inter-observer and intra-observer discrepancies. In this paper, we present an end-to-end deep learning system which directly predicts the H-Score automatically. Our system imitates the pathologists' decision process and uses one fully convolutional network (FCN) to extract all nuclei region (tumour and non-tumour), a second FCN to extract tumour nuclei region, and a multi-column convolutional neural network which takes the outputs of the first two FCNs and the stain intensity description image as input and acts as the high-level decision making mechanism to directly output the H-Score of the input TMA image. To the best of our knowledge, this is the first end-to-end system that takes a TMA image as input and directly outputs a clinical score. We will present experimental results which demonstrate that the H-Scores predicted by our model have very high and statistically significant correlation with experienced pathologists' scores and that the H-Score discrepancy between our algorithm and the pathologists is on par with the inter-subject discrepancy between the pathologists.

Building on crucial insights into the determining factors of the visual integrity of an image and the property of deep convolutional neural network (CNN), we have developed the Deep Feature Consistent Deep Image Transformation (DFC-DIT) framework which unifies challenging one-to-many mapping image processing problems such as image downscaling, decolorization (colour to grayscale conversion) and high dynamic range (HDR) image tone mapping. We train one CNN as a non-linear mapper to transform an input image to an output image following what we term the deep feature consistency principle which is enforced through another pretrained and fixed deep CNN. This is the first work that uses deep learning to solve and unify these three common image processing tasks. We present experimental results to demonstrate the effectiveness of the DFC-DIT technique and its state of the art performances.

In this paper, we introduce deep learning technology to tackle two traditional low-level image processing problems, companding and inverse halftoning. We make two main contributions. First, to the best knowledge of the authors, this is the first work that has successfully developed deep learning based solutions to these two traditional low-level image processing problems. This not only introduces new methods to tackle well-known image processing problems but also demonstrates the power of deep learning in solving traditional signal processing problems. Second, we have developed an effective deep learning algorithm based on insights into the properties of visual quality of images and the internal representation properties of a deep convolutional neural network (CNN). We train a deep CNN as a nonlinear transformation function to map a low bit depth image to higher bit depth or from a halftone image to a continuous tone image. We also employ another pretrained deep CNN as a feature extractor to derive visually important features to construct the objective function for the training of the mapping CNN. We present experimental results to demonstrate the effectiveness of the new deep learning based solutions., Comment: 12 pages

We present a novel method for constructing Variational Autoencoder (VAE). Instead of using pixel-by-pixel loss, we enforce deep feature consistency between the input and the output of a VAE, which ensures the VAE's output to preserve the spatial correlation characteristics of the input, thus leading the output to have a more natural visual appearance and better perceptual quality. Based on recent deep learning works such as style transfer, we employ a pre-trained deep convolutional neural network (CNN) and use its hidden features to define a feature perceptual loss for VAE training. Evaluated on the CelebA face dataset, we show that our model produces better results than other methods in the literature. We also show that our method can produce latent vectors that can capture the semantic information of face expressions and can be used to achieve state-of-the-art performance in facial attribute prediction., Comment: WACV

A popular approach to semantic image understanding is to manually tag images with keywords and then learn a mapping from vi- sual features to keywords. Manually tagging images is a subjective pro- cess and the same or very similar visual contents are often tagged with different keywords. Furthermore, not all tags have the same descriptive power for visual contents and large vocabulary available from natural language could result in a very diverse set of keywords. In this paper, we propose an unsupervised visual theme discovery framework as a better (more compact, efficient and effective) alternative to semantic represen- tation of visual contents. We first show that tag based annotation lacks consistency and compactness for describing visually similar contents. We then learn the visual similarity between tags based on the visual features of the images containing the tags. At the same time, we use a natural language processing technique (word embedding) to measure the seman- tic similarity between tags. Finally, we cluster tags into visual themes based on their visual similarity and semantic similarity measures using a spectral clustering algorithm. We conduct user studies to evaluate the effectiveness and rationality of the visual themes discovered by our unsu- pervised algorithm and obtains promising result. We then design three common computer vision tasks, example based image search, keyword based image search and image labelling to explore potential applica- tion of our visual themes discovery framework. In experiments, visual themes significantly outperforms tags on semantic image understand- ing and achieve state-of-art performance in all three tasks. This again demonstrate the effectiveness and versatility of proposed framework.

We present a method for discovering and exploiting object specific deep learning features and use face detection as a case study. Motivated by the observation that certain convolutional channels of a Convolutional Neural Network (CNN) exhibit object specific responses, we seek to discover and exploit the convolutional channels of a CNN in which neurons are activated by the presence of specific objects in the input image. A method for explicitly fine-tuning a pre-trained CNN to induce an object specific channel (OSC) and systematically identifying it for the human face object has been developed. Based on the basic OSC features, we introduce a multi-resolution approach to constructing robust face heatmaps for fast face detection in unconstrained settings. We show that multi-resolution OSC can be used to develop state of the art face detectors which have the advantage of being simple and compact.

Label ranking aims to learn a mapping from instances to rankings over a finite number of predefined labels. Random forest is a powerful and one of the most successful general-purpose machine learning algorithms of modern times. In this paper, we present a powerful random forest label ranking method which uses random decision trees to retrieve nearest neighbors. We have developed a novel two-step rank aggregation strategy to effectively aggregate neighboring rankings discovered by the random forest into a final predicted ranking. Compared with existing methods, the new random forest method has many advantages including its intrinsically scalable tree data structure, highly parallel-able computational architecture and much superior performance. We present extensive experimental results to demonstrate that our new method achieves the highly competitive performance compared with state-of-the-art methods for datasets with complete ranking and datasets with only partial ranking information., Comment: 28 pages, 4 figures,accepted to Expert Systems With Applications in June 2018

In this paper, we describe a dataset relating to cellular and physical conditions of patients who are operated upon to remove colorectal tumours. This data provides a unique insight into immunological status at the point of tumour removal, tumour classification and post-operative survival. Attempts are made to learn relationships between attributes (physical and immunological) and the resulting tumour stage and survival. Results for conventional machine learning approaches can be considered poor, especially for predicting tumour stages for the most important types of cancer. This poor performance is further investigated and compared with a synthetic, dataset based on the logical exclusive-OR function and it is shown that there is a significant level of 'anti-learning' present in all supervised methods used and this can be explained by the highly dimensional, complex and sparsely representative dataset. For predicting the stage of cancer from the immunological attributes, anti-learning approaches outperform a range of popular algorithms., Comment: IEEE International Conference on Systems, Man, and Cybernetics, pp 797-802, 2012

Bottom-up saliency, an early human visual processing, behaves like binary classification of interest and null hypothesis. Its discriminant power, mutual information of image features and class distribution, is closely related to saliency value by the well-known centre-surround theory. As classification accuracy very much depends on window sizes, the discriminant saliency (power) varies according to sampling scales. Discriminating power estimation in multi-scales framework needs integrating with wavelet transformation and then estimating statistical discrepancy of two consecutive scales (centre-surround windows) by Hidden Markov Tree (HMT) model. Finally, multi-scale discriminant saliency (MDIS) maps are combined by the maximum information rule to synthesize a final saliency map. All MDIS maps are evaluated with standard quantitative tools (NSS,LCC,AUC) on N.Bruce's database with ground truth data as eye-tracking locations ; as well assessed qualitatively by visual examination of individual cases. For evaluating MDIS against well-known AIM saliency method, simulations are needed and described in details with several interesting conclusions, drawn for further research directions.

This paper formulates bottom-up visual saliency as center surround conditional entropy and presents a fast and efficient technique for the computation of such a saliency map. It is shown that the new saliency formulation is consistent with self-information based saliency, decision-theoretic saliency and Bayesian definition of surprises but also faces the same significant computational challenge of estimating probability density in very high dimensional spaces with limited samples. We have developed a fast and efficient nonparametric method to make the practical implementation of these types of saliency maps possible. By aligning pixels from the center and surround regions and treating their location coordinates as random variables, we use a k-d partitioning method to efficiently estimating the center surround conditional entropy. We present experimental results on two publicly available eye tracking still image databases and show that the new technique is competitive with state of the art bottom-up saliency computational methods. We have also extended the technique to compute spatiotemporal visual saliency of video and evaluate the bottom-up spatiotemporal saliency against eye tracking data on a video taken onboard a moving vehicle with the driver's eye being tracked by a head mounted eye-tracker.

The bottom-up saliency, an early stage of humans' visual attention, can be considered as a binary classification problem between centre and surround classes. Discriminant power of features for the classification is measured as mutual information between distributions of image features and corresponding classes . As the estimated discrepancy very much depends on considered scale level, multi-scale structure and discriminant power are integrated by employing discrete wavelet features and Hidden Markov Tree (HMT). With wavelet coefficients and Hidden Markov Tree parameters, quad-tree like label structures are constructed and utilized in maximum a posterior probability (MAP) of hidden class variables at corresponding dyadic sub-squares. Then, a saliency value for each square block at each scale level is computed with discriminant power principle. Finally, across multiple scales is integrated the final saliency map by an information maximization rule. Both standard quantitative tools such as NSS, LCC, AUC and qualitative assessments are used for evaluating the proposed multi-scale discriminant saliency (MDIS) method against the well-know information based approach AIM on its released image collection with eye-tracking data. Simulation results are presented and analysed to verify the validity of MDIS as well as point out its limitation for further research direction., Comment: arXiv admin note: substantial text overlap with arXiv:1301.3964