We are always looking for motivated people in all levels (undergraduate, masters, PhD, post-doc). Interested are highly welcome to contact us with a brief CV to discuss the details oriented on their interests. Contact
Projects
Light Field Super-resolution
Light Field Imaging is based on capturing the light rays in different location and angles by using Light Field Cameras. This technique presents angular resolution and spatial resolution. However, these are not enough to obtain better performance in post-capture refocusing, controlling and computing depth-of-field. In this study, the development of a unique patch-based Light Field Super-resolution technique is aimed.
Denoising for Fluorescence Microscopy Images
As noise corruption is an inevitable issue for all imaging technologies, this problem causes serious difficulties in analyzing the biological fine-details of fluorescence microscopy images. While Gaussian only, Poisson only and mixture of Poisson-Gaussian can generally be observed, the mixed-noise is more prominent in fluorescence microscopy. In this paper, a novel patch-based denoiser-learning approach is proposed for the images captured by fluorescence microscopy. The developed algorithm mainly builds upon linear-embeddings of neighboring image patches, and it learns a linear transformation between noisy and clean intrinsic geometric properties of patch-spaces. Experimental results demonstrate that the proposed “Neighbor Linear-Embedding Denoising” (NLED) has competitive performance both visually and statistically when compared to other algorithms in literature, for noise corrupted fluorescence microscopy images.
Multi Exposure Image Fusion Using Sparse Representation
The rising popularity of High Dynamic Range (HDR) imaging combined with regular Low Dynamic Range (LDR) display devices created a problem of dynamic range gap. This gap caused different types of distortions and more importantly data loss while displaying HDR images in an LDR display device. Thus, a need for a solution rose. Multi Exposure Fusion (MEF) aims to solve this problem by fusing images with different exposure levels. In this project we propose a novel MEF algorithm using sparse representations of the input images.
Smartphone-based automated optofluidic platform for cell counting applications
The number of cells is important in order to seed cells for subsequent experiments, monitor cell health and proliferation rates, assess immortalization or transformation, detect transfection or infection, and arranged cell-based assays. Manual cell counting with a hemocytometer is the most common method recently. Manual cell counting is a low-cost method, but it can take a long time depending on the number of samples. Furthermore, the experiment may have a negative impact depending on the risk of the laboratory specialist performing the count making a mistake. In this project, it is aimed to calculate cell viability and cell confluency using a smartphone-based automated optofluidic platform. In this platform, cell-based imaging will be performed with optical magnification and pixel-based digital magnification with a lens integrated into the front of the phone camera. In this way, the platform will be portable and affordable. Also, the pre-processing of cells for cell counting is also a time-consuming step. A microfluidic system will be designed that will automatically transfer the solutions used in cell preparation into the cells within the polydimethylsiloxane (PDMS) based microfluidic chip in order to solve this problem while also speeding up the process. The detection of cell viability by the platform with the image processing algorithm will increase the accuracy of the device by minimizing user errors in the process. A Java-based mobile application will be designed to take cell photos and show the user the results of the counting.
Facial Texture and Emotion/Expression Transfer
Novel algorithms are investigated for transfering facial texture, emotional state (such as surprise, smile, anger, fear), and facial features between human faces through sparse and redundant representations. The facial morphing via Delaunay Triangulation and Voronoi Diagram, face detection, and facial landmarks detection are carried out in pre-processing steps. The success rate and quality of output images are assessed via Frechet Inception Distance (FID), Learned Perceptual Image Patch Similarity (LPIPS) scores, and deep-learning based facial feature classifier model.
Development of Mobile Application that carries out the Transformation of Sign Language and Audio Signals to Texts
The communication problems of hearing-impaired individuals in social life and the inadequacy and high cost of solutions to these problems are known by everyone. Our goal at the beginning of this project was to reintegrate hearing impaired individuals into society by producing a more effective, convenient and cost-effective solution. In line with this purpose, we have been developing a mobile application that will be integrated into iOS and Android platforms by converting sign language and audio signals into text format, where two-way communication will be carried out effectively.
Microscale Image Enhancement
The restrictions of accessing high-end microscopes, microscale cameras and high-tech imaging lenses result in a high demand on low-cost microscopes. However, low-cost microscopes are facing with many image capture and quality limitations due to incompatible equipped instrumentation. This study aims at overcoming illumination and contrast problems, color aberration issues, and blur and noise corruption in low-cost microscopes at high image magnification rates. The three color channels of the input image are enhanced via principal component analysis and well-exposedness feature maps by means of cross-channel histogram matching, Laplacian and non-local means filtering. The proposed approach produces sharper, and better color and illumination fixed outputs when compared to existing methods in literature.
Underwater Image Dehazing
Due to light scattering and absorption while traveling through water, underwater images become hazy and loose critical information resulting in poor contrast and color performance. Hence, it is difficult to see the difference between foreground colors and items, and to differentiate the background in these images. To solve these issues, this study proposes a novel technique for single underwater image enhancement which is based on the recovery of the lost red-channel via a weighted multi-scale fusion. Firstly, three color balance algorithms are applied to the input image to gain more information about the scene. Then, five weight maps are extracted from these balanced versions of the input image to emphasize the details. Finally, the enhanced output is obtained with the new red-channel, white balanced green and blue channels followed by gamma correction to maintain contrast of the image. The developed method produces higher-quality underwater images that can be evaluated qualitatively and quantitatively when compared to state-of-the-art approaches.
Image Quality Assessment Based On Manifold Distortion
An image quality metric is proposed by introducing a new framework for full reference image quality assessment from the perspective of image patch manifolds. Assuming that most natural scenes are sampled from low dimensional manifolds or submanifolds, perceived image degradations in structural variations can be quantitatively evaluated on the surfaces of highly nonlinear image manifolds. Manifold distortion image quality index first characterizes intrinsic geometric properties of the locally linear manifold structures of spatially local patch spaces, and then measures the deviation from the original smooth manifold structure to calculate the distortion index. Experimental results demonstrate a strong promise with a comparison to both subjective evaluation and state-of-the-art objective quality assessment methods.
Multi-Exposure Image Fusion
The visual system enables humans to perceive all details of the real-world with vivid colors, while high dynamic range (HDR) technology aims at capturing natural scenes in a closer way to human perception through a large dynamic range of color gamut. Especially for traditional -low dynamic range (LDR)- devices, HDR-like image generation is an attractive research topic. Blending a stack of input LDR exposures is called multi-exposure image fusion (MEF). MEF is indeed a very challenging problem and it is highly prone to halo effects or ghosting and motion blur in the cases when there are spatial discontinuities in between input exposures. To overcome these artifacts, MEF keeps the "best" quality regions of each exposure via a weight characterization scheme. This paper proposes an effective weight map extraction framework which relies on principal component analysis, adaptive well-exposedness and saliency maps. The characterized maps are later refined by a guided filter and a blended output image is obtained via pyramidal decomposition. Comprehensive experiments and comparisons demonstrate that the developed algorithm generates very strong statistical and visual results for both static and dynamic scenes. In addition, the designed method is successfully applied to the visible-infrared image fusion problem without any further optimization.
Melanoma Detection in Dermoscopic Images
Melanoma is a skin cancer caused by the ultraviolet radiation from the sun. If it is not detected at early stages, melanoma will become severe and more importantly it may spread to the other body organs, most commonly to the lungs, brain, liver and bones. Dermatologists look for the tell tales on the pigmented lesions (moles) on the skin to detect melanoma, or for some cases discriminate it from other skin diseases. Unfortunately, imprecise subjective analysis may result in the form of a series of biopsies which maybe not needed. Furthermore, this type of imprecision may allow a melanoma case to grow without a notice. To overcome this challenge, an automatic melanoma detection system is proposed in this study. The developed approach is based on Bag of Visual Words (BoVW) which includes both traditional and new age methods. Experimental comparisons between this novel approach and well-known convolutional neural network models show the effectiveness of the proposed model.
Enhancing Two-Photon Images for Anatomical Visualisation
Two-photon Laser Scanning Microscopy (2P-LSM) is a technique used to image the living tissue with relatively high spatio-temporal resolution. However, the time-series images are often corrupted with Poisson-Gaussian noise and deteriorated with motion artifacts. This paper deals with the problem of enhancing 2P-LSM images to reconstruct high quality and high spatial-resolution outputs using the observed time-series stack of low-resolution images. The proposed technique consists of several components including noise filtering, image registration, cell detection and focus measurement, and clustering for a joint denoising and super-resolution. Extensive experiments demonstrate that the proposed method results in gratifying output images containing apparent and clear cell forms at different focus levels.
Detecting Bleach Quality Of Denim Products
In order for denim to obtain desired color tones and reach our hands in its many forms, it needs to go through the process of bleaching. This can be done by operations such as stone washing or ozone finishing. After this process, whether the denim is bleached to the desired color with minimal mistake is decided by workers. In this study, we aim to develop algorithms that will eliminate the human factor and decide on its own whether the denim is bleached correctly and its probability.
Image Denoising through Neighbor-Embedding
The main purpose of this study is to develop an algorithm through neighbor-embedding for blind image denoising. Similar patches from the original image and down-scaled versions are collected to create a dictionary. An iterative local-embedding algorithm with an error threshold level is employed to calculate estimated noise-free patches. The denoised image is obtained by getting weighted averaging of overlapping estimated patches.
Audio-Visual Lip Reading
Lip reading, described as extracting speech data from the observable deeds in the face, particularly the jaws, lips, tongue and teeth, is a very challenging task. It is indeed a beneficial skill that helps people to comprehend and interpret the content of other people’s speech, when it is not sufficient to recognize either audio or expression. Even experts require a certain level of experience and need an understanding of visual expressions to interpret spoken words. However, this may not be efficient enough for the process. Nowadays, lip sequences can be converted into expressive words and phrases with the aid of computers. Thus, the usage of neural networks (NNs) is increased rapidly in this field. The main contribution of this study is to use Short-Time Fourier Transformed (STFT) audio data as an extra image input and employing 3D Convolutional NNs (CNNs) for feature extraction. This generates features considering the change in consecutive frames and makes use of visual and auditory data together with the attributes from the image. After testing several experimental scenarios, it turns out to be the proposed method has a strong promise for further development in this research domain.
Food Quality Assessment
In this study, a meat quality assessment algorithm based on deep leaning is designed to be used in the meat section of a supermarket chain. It is aimed to distinguish between fresh and spoiled food, and to identify meat that will be spoiled in a short time to put it on sale and reduce the economic loss. As an extension to this study, a fish freshness classification study is carried out, and a dataset is formed by images of nine different seafood types captured in the fish section of a supermarket. To the best of available knowledge, this is the first large-scale fish dataset consisting of widely consumed fish. In an additional fruit freshness study, segmentation is carried out for the fruit types and then different feature extraction methods are compared for the classification of fresh and rotten bananas, apples, and oranges.
Electronic-Nose Systems: Applications and Comparative Analysis
In the last two decades, improvements in materials, sensors and machine learning technologies have led to a rapid extension of electronic nose (EN) related research topics with diverse applications. The food and beverage industry, agriculture and forestry, medicine and health-care, indoor and outdoor monitoring, military and civilian security systems are the leading fields which take great advantage from the rapidity, stability, portability and compactness of ENs. Although the EN technology provides numerous benefits, further enhancements in both hardware and software components are necessary for utilizing ENs in practice. This paper provides an extensive survey of the EN technology and its wide range of application fields, through a comprehensive analysis of algorithms proposed in the literature, while exploiting related domains with possible future suggestions for this research topic.
Image-Declipping: Saturation Correction in Single Images
High dynamic range (HDR) images present fine details in a scene and are visually more appealing than low dynamic range (LDR) images, since they contain a greater dynamic range of color gamut. HDR compatible displays are currently high-cost, therefore tone-mapping algorithms have widely been used to obtain high quality images for LDR screens with a lower cost. However, tone-mapped images may contain clipped pixel regions, which should be corrected to retrieve the lost information, to acquire visually pleasing LDR images. In a single image, the recovery of color and texture information in clipped regions is challenging, yet an attractive research field in image processing. Although there are several algorithms present in literature, developing a general framework for different types of image content is hard to achieve. This study proposes a single image declipping method based on linear embeddings, difference of pixels and block-search. Experimental results carried out on a tone-mapped HDR image dataset and LDR images demonstrate that the proposed algorithm is able to successfully recover saturated pixels in various types of images. Detailed statistical and visual comparisons show that this approach produces superior results on average for both tone-mapped and LDR images when compared to existing techniques.
Binocular Vision based Convolutional Networks
It is arguable that whether the single camera captured (monocular) image datasets are sufficient enough to train and test convolutional neural networks (CNNs) for imitating the biological neural network structures of the human brain. As human visual system works in binocular, the collaboration of the eyes with the two brain lobes needs more investigation for improvements in such CNN-based visual imagery analysis applications. It is indeed questionable that if respective visual fields of each eye and the associated brain lobes are responsible for different learning abilities of the same scene. There are such open questions in this field of research which need rigorous investigation in order to further understand the nature of the human visual system, hence improve the currently available deep learning applications. This paper analyses a binocular CNNs architecture that is more analogous to the biological structure of the human visual system than the conventional deep learning techniques. While taking a structure called optic chiasma into account, this architecture consists of basically two parallel CNN structures associated with each visual field and the brain lobe, fully connected later possibly as in the primary visual cortex. Experimental results demonstrate that binocular learning of two different visual fields leads to better classification rates on average, when compared to classical CNN architectures.
