Group Of Pixels Research Articles

Statistically optimized inversion algorithm for enhanced retrieval of aerosol properties from spectral multi-angle polarimetric satellite observations

Abstract. The proposed development is an attempt to enhance aerosol retrieval by emphasizing statistical optimization in inversion of advanced satellite observations. This optimization concept improves retrieval accuracy relying on the knowledge of measurement error distribution. Efficient application of such optimization requires pronounced data redundancy (excess of the measurements number over number of unknowns) that is not common in satellite observations. The POLDER imager on board the PARASOL micro-satellite registers spectral polarimetric characteristics of the reflected atmospheric radiation at up to 16 viewing directions over each observed pixel. The completeness of such observations is notably higher than for most currently operating passive satellite aerosol sensors. This provides an opportunity for profound utilization of statistical optimization principles in satellite data inversion. The proposed retrieval scheme is designed as statistically optimized multi-variable fitting of all available angular observations obtained by the POLDER sensor in the window spectral channels where absorption by gas is minimal. The total number of such observations by PARASOL always exceeds a hundred over each pixel and the statistical optimization concept promises to be efficient even if the algorithm retrieves several tens of aerosol parameters. Based on this idea, the proposed algorithm uses a large number of unknowns and is aimed at retrieval of extended set of parameters affecting measured radiation. The algorithm is designed to retrieve complete aerosol properties globally. Over land, the algorithm retrieves the parameters of underlying surface simultaneously with aerosol. In all situations, the approach is anticipated to achieve a robust retrieval of complete aerosol properties including information about aerosol particle sizes, shape, absorption and composition (refractive index). In order to achieve reliable retrieval from PARASOL observations even over very reflective desert surfaces, the algorithm was designed as simultaneous inversion of a large group of pixels within one or several images. Such multi-pixel retrieval regime takes advantage of known limitations on spatial and temporal variability in both aerosol and surface properties. Specifically the variations of the retrieved parameters horizontally from pixel-to-pixel and/or temporary from day-to-day are enforced to be smooth by additional a priori constraints. This concept is expected to provide satellite retrieval of higher consistency, because the retrieval over each single pixel will be benefiting from coincident aerosol information from neighboring pixels, as well, from the information about surface reflectance (over land) obtained in preceding and consequent observations over the same pixel. The paper provides in depth description of the proposed inversion concept, illustrates the algorithm performance by a series of numerical tests and presents the examples of preliminary retrieval results obtained from actual PARASOL observations. It should be noted that many aspects of the described algorithm design considerably benefited from experience accumulated in the preceding effort on developments of currently operating AERONET and PARASOL retrievals, as well as several core software components were inherited from those earlier algorithms.

Processing MUSE hyperspectral data: Denoising, deconvolution and detection of astrophysical sources

We address the problem of processing astronomical hyperspectral data cubes in the context of the forthcoming MUSE instrument. MUSE, which is under construction, will provide massive hyperspectral data with about 300 × 300 pixels at approximately 4000 wavelengths. One of its main astrophysical objectives concerns the observation of extragalactic deep fields, where MUSE should be able to detect and characterize galaxies much fainter than the ones currently observed by other ground-based instruments. The data will suffer, however, from very powerful and spectrally variable perturbations. In this paper, MUSE data cubes are first considered as a collection of spectra, which are processed independently. A restoration method is proposed, based on the hypothesis that data can be approximated by appropriate sparse representations. Sparsity can be naturally expressed in the spectral domain, where a galaxy spectrum is mainly the superposition of an emission and absorption line spectrum, which is naturally sparse, on a continuum, which is supposed to have a sparse discrete cosine transform. The problem is addressed within the ℓ 1 -norm penalization setting. The original features of the model consist, first, in taking into account observational specificities such as the spectrally variable instrumental response and non-identically distributed noise, and, second, in tuning regularization parameters in this setting, which are fixed in order to obtain uniform false alarm rates for decomposition coefficients. In a second step, such sparse decompositions are used as an input to an object detection and characterization method. The decomposed spectra are first used for spatial segmentation. Then, once a group of pixels has been identified as belonging to the same object, the corresponding spectrum and amplitude map are jointly estimated under the former sparsity assumption. Applications to object identification, the amplitude map and spectrum estimation are presented for realistic deep field simulated data cubes provided by the MUSE consortium.

Image-space hierarchical coherence buffer

Indirect illumination plays an important role in global illumination. However, computing indirect illumination is a time-consuming process and needs to be approximated to achieve interactive performance. Indirect illumination varies rather slowly across the surface. This leads to the idea of computing indirect illumination sparsely in the scene and interpolating the result. This paper presents a hierarchical structure, which enables efficient sampling. The hierarchy is constructed in the image space by exploiting coherences among the screen-space pixels. From the hierarchy, samples are chosen, each of which represents a group of coherent pixels. This paper presents two methods of utilizing the samples for indirect lighting computation. The methods produce plausible lighting results and show high performances. The proposed algorithms run entirely in the image space and are easy to implement in contemporary graphic hardware.

Automatic segmentation of tissue sections using the multielement information provided by LA-ICP-MS imaging and k-means cluster analysis

Laser Ablation Inductively Coupled Plasma Mass Spectrometry (LA-ICP-MS) is an established and powerful tool to analyse the distribution of elements in tissue sections. Among other applications, the technique is expected to play a central role in the understanding of normal and pathological element distributions in brain tissue. In order to interpret the distribution of elements such as the bio-metals Cu, Zn, Fe and Mn and proceed to an element-based comparison between groups of samples, it is necessary to anatomically parcel the tissue section into regions-of-interest and to average element signals across these regions. This categorization, also termed segmentation, can be done manually, but the support of automated procedures is highly desirable, especially in order to (1) identify groups of pixels with similar elemental fingerprint, termed clusters, and to determine which degree of discrimination is reasonable; (2) segment anatomical structures known to exhibit substructure but without clearly defined borders, such as the healthy cortex, zones of tumours or ischemic lesions, in an observer-independent way; and (3) to investigate correlation between the distribution of elements in tissue and phenomena which incorporate contributions from several elements in a convoluted way, such as the origin of contrast in magnetic resonance imaging (MRI) experiments. The multi-parametric information provided by LA-ICP-MS lends itself naturally to multivariate analysis. This study provides a new way to synthesise the information distributed over many element images by demonstrating the possibility to segment tissue sections into biologically meaningful substructures. This data-driven, observer-independent categorization was based on k-means clustering. The optimal number of clusters was determined based on the silhouette method. Segmentation of healthy tissue resulted in a set of substructures in perfect congruence to the anatomical architecture. Segmentation of ischemic lesions identified a number of regions with different fingerprints of C, P, Fe, Cu and Zn deposits. Clustering provides a promising way of combining the information present in several element images and reveals structure which is not entirely present in any isolated image. As a useful by-product of this study we have found a promising method for investigating the optimal line length within the process of image reconstruction from the continuous stream of raw data points. Images were characterized by their tensor of inertia, in image- as well as in Fourier dual-space ( k-space) and changes in the ratio of the intrinsic moments of inertia or the orientation of the principal axes were found to closely describe the optimum orientation. The first results look very encouraging, but the method must be extensively tested before it can be used as an automatic procedure. In conclusion, cluster analysis of mass spectrometric imaging data allows one to define the fingerprint element distribution of different anatomically or functionally distinct regions and opens a new way for the study of correlation between the element distribution and related phenomena.

Adaptive Marginal Median Filter for Colour Images

This paper describes a new filter for impulse noise reduction in colour images which is aimed at improving the noise reduction capability of the classical vector median filter. The filter is inspired by the application of a vector marginal median filtering process over a selected group of pixels in each filtering window. This selection, which is based on the vector median, along with the application of the marginal median operation constitutes an adaptive process that leads to a more robust filter design. Also, the proposed method is able to process colour images without introducing colour artifacts. Experimental results show that the images filtered with the proposed method contain less noisy pixels than those obtained through the vector median filter.

Mode estimation in high-dimensional spaces with flat-top kernels: Application to image denoising

Mode estimation is extensively studied in statistics. One of the most widely used methods of mode estimation is hill-climbing on a kernel density estimator with gradient ascent or a fixed-point approach. Within this framework, Gaussian kernels proves to be a natural and intuitive option for non-parametric density estimation. This paper shows that in the case of high-dimensional data, mode estimation can be improved by using differently shaped kernels, called flat-top kernels. The improvement are illustrated with an image denoising application, in which pictures are decomposed into small patches, i.e. groups of adjacent pixels, that are vectorized. Noise in the patches can be attenuated by substituting them with the closest mode in the observed distribution of patches. The quality of the denoised picture then depends on the accuracy of mode estimation in a high-dimensional space. Experiments conducted on usual benchmarks in the image processing community show that flat-top kernels outperform the Gaussian one.

Super multi-view display with a lower resolution flat-panel display

A lenticular-type super multi-view (SMV) display normally requires an ultra high-resolution flat-panel display. To reduce this resolution requirement, two or more views are generated around each eye with an interval smaller than the pupil diameter. Cylindrical lenses constituting a lenticular lens project a group of pixels of the flat-panel display to generate a group of viewing zones. Pixel groups generating left and right viewing zones through the same cylindrical lens are partitioned to separate the two zones. The left and right pixel groups for different cylindrical lenses are interlaced horizontally. A prototype SMV display is demonstrated.

Detection of pore space in CT soil images using artificial neural networks

Abstract. Computed Tomography (CT) images provide a non-invasive alternative for observing soil structures, particularly pore space. Pore space in soil data indicates empty or free space in the sense that no material is present there except fluids such as air, water, and gas. Fluid transport depends on where pore spaces are located in the soil, and for this reason, it is important to identify pore zones. The low contrast between soil and pore space in CT images presents a problem with respect to pore quantification. In this paper, we present a methodology that integrates image processing, clustering techniques and artificial neural networks, in order to classify pore space in soil images. Image processing was used for the feature extraction of images. Three clustering algorithms were implemented (K-means, Fuzzy C-means, and Self Organising Maps) to segment images. The objective of clustering process is to find pixel groups of a similar grey level intensity and to organise them into more or less homogeneous groups. The segmented images are used for test a classifier. An Artificial Neural Network is characterised by a great degree of modularity and flexibility, and it is very efficient for large-scale and generic pattern recognition applications. For these reasons, an Artificial Neural Network was used to classify soil images into two classes (pore space and solid soil). Our methodology shows an alternative way to detect solid soil and pore space in CT images. The percentages of correct classifications of pore space of the total number of classifications among the tested images were 97.01%, 96.47% and 96.12%.

Per-pixel vs. object-based classification of urban land cover extraction using high spatial resolution imagery

In using traditional digital classification algorithms, a researcher typically encounters serious issues in identifying urban land cover classes employing high resolution data. A normal approach is to use spectral information alone and ignore spatial information and a group of pixels that need to be considered together as an object. We used QuickBird image data over a central region in the city of Phoenix, Arizona to examine if an object-based classifier can accurately identify urban classes. To demonstrate if spectral information alone is practical in urban classification, we used spectra of the selected classes from randomly selected points to examine if they can be effectively discriminated. The overall accuracy based on spectral information alone reached only about 63.33%. We employed five different classification procedures with the object-based paradigm that separates spatially and spectrally similar pixels at different scales. The classifiers to assign land covers to segmented objects used in the study include membership functions and the nearest neighbor classifier. The object-based classifier achieved a high overall accuracy (90.40%), whereas the most commonly used decision rule, namely maximum likelihood classifier, produced a lower overall accuracy (67.60%). This study demonstrates that the object-based classifier is a significantly better approach than the classical per-pixel classifiers. Further, this study reviews application of different parameters for segmentation and classification, combined use of composite and original bands, selection of different scale levels, and choice of classifiers. Strengths and weaknesses of the object-based prototype are presented and we provide suggestions to avoid or minimize uncertainties and limitations associated with the approach.

Agglomerative connectivity constrained clustering for image segmentation

AbstractWe consider the problem of clustering under the constraint that data points in the same cluster are connected according to a pre‐existed graph. This constraint can be efficiently addressed by an agglomerative clustering approach, which we exploit to construct a new fully automatic segmentation algorithm for color photographs. For image segmentation, if the pixel grid with eight neighbor connectivity is imposed as the graph, each group of pixels generated by this clustering method is ensured to be a geometrically connected region in the image, a desirable trait for many subsequent operations. To achieve scalability for images with large sizes, the segmentation algorithm combines the top‐down k‐means clustering with the bottom‐up agglomerative clustering method. We also find that it is advantageous to conduct clustering at multiple stages through which the similarity measure is adjusted. Experimental results with comparison to other widely used and state‐of‐the‐art segmentation methods show that the new algorithm achieves higher accuracy at much faster speed. A software package is provided for public access. © 2011 Wiley Periodicals, Inc. Statistical Analysis and Data Mining 4: 84–99 2011

A Novel Asynchronous Pixel for an Energy Harvesting CMOS Image Sensor

This paper proposes a novel energy harvesting technique based on an asynchronous pixel structure and an efficient energy generation scheme, referred to as avalanche energy generation (AEG). The key idea behind using an asynchronous type of pixel is to lower the power consumption by enabling only active pixels to be read-out after which they enter into a power generation mode. In this mode, the on-pixel photodetector itself will be used to harvest the light energy from the environment and make it available to active pixels. A very interesting feature about our proposed approach is that during a frame capture, critical energy is mainly required for starting-up activity. Once a group of pixels have been read-out, the available energy will rise and more array activity will contribute to the generation of more energy, hence creating an avalanche effect. In contrast to other early designs of energy harvesting image sensors, our scheme uses the photodetector itself for power generation. This results in better utilization of the photosensitive area and more importantly an improved energy generation scheme. Detailed power analysis and extensive simulation results are provided in this paper, which validate the proposed concept. Three test structures have been fabricated in AMIS 1-poly, 5-metal CMOS 0.35-m n-well process. The power generation process and event generation have been successfully verified experimentally.

4+1-transistor pixel architecture for high-speed, high-resolution CMOS image sensors

A pixel architecture is introduced which allows a drastic reduction of the column capacitance of a monolithic pixel array. It consists of a classic 4T pixel architecture together with an extra switch added at regular positions in the column array and shared by a group of pixels of the column. In this way, each pixel will see an output capacitance proportional to the number of pixels sharing the extra switch and the total number of extra switches.

Object-based image analysis of remote sensing data

Remote sensing has developed various methods and technologies for contactless and cost-effective mapping of large area land cover/land use maps and other thematic maps. The key factor for the availability and reliability of these maps for use in Earth sciences is the development of effective procedures for satellite data analysis and classification. The most appropriate approach for classifying low and medium resolution satellite images (pixel size is coarser than, or at best similar to, the size of geographical objects) is pixelbased classification in which an individual pixel is classified into the closest class based on its spectral similarity. With increasing spatial resolution, pixel-based classification methods became less effective, since the relationship between the pixel size and the dimension of the observed objects on the Earth's surface has changed significantly. Therefore object-oriented classification has become increasingly popular over the past decade. This combines segmentation (which is a fundamental phase of the approach) and contextual classification. Segmentation divides the image into homogeneous pixel groups (segments), which are – during the semantic classification process arranged into classes based on their spectral, geometric, textural and other features during. The intent of this paper is to present the theoretical argumentation and methodology of object-based image analysis of remote sensing data, provide an overview of the field and point out certain restrictions as regards the current operational solutions.

A Cluster-Analysis-Based Efficient Multibaseline Phase-Unwrapping Algorithm

Multibaseline phase unwrapping is a critical processing procedure of multibaseline synthetic aperture radar interferometry (InSAR). It has an advantage over single-baseline phase unwrapping in discontinuous-terrain-height estimation. In this paper, the pixels' combination information of multiple InSAR interferograms with different baseline lengths is deeply investigated, and the term ambiguity vector is proposed to represent a pixel's ambiguity number of multiple interferograms. It is also revealed that pixels with the same ambiguity vector have an exclusive pattern among them. A fast cluster-analysis (CA)-based method is proposed for multibaseline phase unwrapping. In this method, all pixels are first clustered into different groups according to their patterns, and then, information of the cluster center is used to unwrap the phases of pixels group by group. Simulation results are shown to verify the effectiveness, efficiency, and noise robustness of CA multibaseline phase-unwrapping method.

Adaptive imaging system with spatial light modulator for robust shape measurement of partially specular objects

In imaging systems, when specular surfaces responding sensitively to varying illumination conditions are imaged on groups of CCD pixels using imaging optics, the obtained image usually suffers from pixel saturation, resulting in smearing or blooming phenomena. These problems are then serious obstacles when applying structured light-based optical profiling methods to the shape measurement of general objects with partially specular surfaces. Therefore, this paper combines a phase-based profiling system with an with an adaptive spatial light modulator in the imaging part for measuring the three-dimensional shapes of objects with an advanced dynamic range. The use of a spatial light modulator in front of a CCD camera prevents the image sensor from being saturated, as the pixel transmittance is controlled by monitoring the input images and providing modulator feedback signals over time and space. When using the proposed system, since the projected fringes are effectively imaged on the CCD without any pixel saturation, phase information according to the object's shape can be correctly extracted from non-saturated images. The configuration of the proposed system and transmittance control scheme are explained in detail, plus the performance is verified through a series of experiments, in which phase information was successfully extracted from areas that are not normally measurable due to saturation. Based on the results, the proposed shape measurement system showed a more advanced adaptive dynamic range when compared with a conventional system.

Tropical forest phenology in French Guiana from MODIS time series

Despite a growing need for knowledge on how terrestrial ecosystems function, phenological studies are rather rare. This study characterizes the temporal evolution of forest covers in French Guiana using variations in the Moderate Resolution Imaging Spectroradiometer (MODIS) satellite signal and vegetation indices, such as the Shortwave Infrared Water Stress Index and the Enhanced Vegetation Index. The principle consists of reconstructing a temporal cycle mosaiced from a 9-year database. Rather than extract the phenological information from each pixel, dynamics were extracted from homogeneous groups of forest pixels enabling the identification of different phenological phases in forest ecosystems. A north–south gradient of chlorophyll activity emerged. The further we moved to the south, the more forest groups consist of fast-growing and drought-tolerant species, whereas a north–south shift appeared at the start dates of phenological cycles. This is a promising approach that paves the way for a much deeper understanding of the mechanisms governing the rainforest in French Guiana, as well as the responses of the Amazon forest to climate change and drought in the long term.

Bacterial mixture identification using Raman and surface‐enhanced Raman chemical imaging

AbstractThe ability of normal Raman and surface‐enhanced Raman scattering (SERS) to identify and detect bacteria has shown great success in recent studies. The addition of silver nanoparticles to bacterial samples not only results in an enhanced Raman signal, but it also suppresses the native fluorescence associated with biological material. In this report, Raman chemical imaging (RCI) was used to analyze individual bacteria and complex mixtures of spores and vegetative cells. RCI uses every pixel or a binned pixel group (BPG) of the Raman camera as an independent Raman spectrograph, allowing collection of spatially resolved Raman spectra. The advantage of this technique resides primarily in the analysis of samples in complex backgrounds without the need for physically isolating or purifying the sample. Using a chemical imaging Raman microscope, we compare normal RCI to SERS‐assisted chemical imaging of mixtures of bacteria. In both cases, we are able to differentiate single bacterium in the Raman microscope's field of view, with a 60‐fold reduction in image acquisition time and a factor of 10 increase in the signal‐to‐noise ratio for SERS chemical imaging over normal RCI. Copyright © 2010 John Wiley & Sons, Ltd.

Assessing a Template Matching Approach for Tree Height and Position Extraction from Lidar-Derived Canopy Height Models of Pinus Pinaster Stands

In this paper, an assessment of a method using a correlation filter over a lidar-derived digital canopy height model (CHM) is presented. The objective of the procedure is to obtain stem density, position, and height values, on a stand with the following characteristics: ellipsoidal canopy shape (Pinus pinaster), even-aged and single-layer structure. The process consists of three steps: extracting a correlation map from CHM by applying a template whose size and shape resembles the canopy to be detected, applying a threshold mask to the correlation map to keep a subset of candidate-pixels, and then applying a local maximum filter to the remaining pixel groups. The method performs satisfactorily considering the experimental conditions. The mean tree extraction percentage is 65% with a coefficient of agreement of 0.4. The mean absolute error of height is ~0.5 m for all plots except one. It can be considered a valid approach for extracting tree density and height in regularly spaced stands (i.e., poplar plantations) which are fundamental for extracting related forest parameters such as volume and biomass.

Optical simulation for cross‐talk evaluation and improvement of autostereoscopic 3‐D displays with a projector array

Abstract— Multi‐view spatial‐multiplexed autostereoscopic 3‐D displays normally use a 2‐D image source and divide the pixels to generate perspective images. Due to the reduction in the resolution of each perspective image for a large view number, a super‐high‐resolution 2‐D image source is required to achieve 3‐D image quality close to the standard of natural vision. This paper proposes an approach by tiling multiple projection images with a low magnification ratio from a microdisplay to resolve the resolution issue. Placing a lenticular array in front of the tiled projection image can lead to an autostereoscopic display. Image distortion and cross‐talk issues resulting from the projection lens and pixel structure of the microdisplay have been addressed with proper selection of the active pixel and adequate pixel grouping and masking. Optical simulation has shown that a 37‐in. 12‐view autostereoscopic display with a full‐HD (1920 × 1080) resolution can be achieved with the proposed 3‐D architecture.

An adaptive region based color texture segmentation using fuzzified distance metric

Texture is an important spatial feature, useful for identifying object or region of interest. In texture analysis the foremost task is to extract texture features, which efficiently embody the information about the textural characteristics of the image. This can be used for the segmentation of different texture images. Texture segmentation is the process of partitioning an image into regions with different textures containing similar group of pixels. This paper, mainly focus on the unsupervised segmentation of color textured images as a prototypical application in computer vision. Here the objective is to group pixels or small image patches such that meaningful regions of identical/similar texture are obtained. This paper presents a new approach for color texture segmentation using Haralick features extracted from color co-occurrence matrices. The originality of this approach is to select the most discriminating color texture features extracted from the color co-occurrence matrices calculated in HSI color space. Fuzzified distance metric is used for achieving texture segmentation.