Image and Signal Processing

When tracking an object, it is reasonable to assume that the dynamic model can change over time. In practical applications, Interacting Multiple Model (IMM) filter are a popular choice for considering such varying system characteristics. The motion of the object is often modeled using position, velocity, and acceleration. It seems obvious that different image space dimensions can be considered in one overall system state vector. In this paper, the fallacy of simply extending the state vector in case of tracking an object solely in image space is demonstrated. Thereby, we show how under such conditions the effectiveness of an IMM filter can be improved by separating particular states. The proposed approach is evaluated on the VOT 2014 dataset.

Local properties of image (phase, amplitude and orientation) can be estimated practically using quadrature filters kernel and can be easily represented in two dimensions using the monogenic signal. This powerful feature representation has given rise to robust phase-based edge detection. Nonetheless, it is limited to the class of intrinsically one-dimensional signals, such as lines and edges. All other possible local patterns such as corners and junction are of intrinsic dimension two. Our aim in this paper is to present a new edge detection method for extracting local features of any curved signal. It is based on the conformal monogenic signal which is in practical applications compatible with intrinsically one and two-dimensional signal. Using different filters, our model have been tested and compared with classical models and some recent ones. The preliminary results show that our detection technique is more efficient and more accurate.

In this paper, we present a new way of 2D feature extraction. We start by showing the direct link that exist between the Riesz Transform (RT) and the gradient and Laplacian operators. This formulation allows us to interpret the RT as a gradient of a smoothed image. Thus, by analogy with the classical models, the maximum gradient and the zero crossings of the divergence of the TR provide information about the position of contours. The interest of the RT is its representation that naturally sweeps the whole area of the image and allows a correct description of structures. Using different filters, our models have been tested and compared with classical models and some recent ones. The results show that our detection technique is more efficient and more accurate.

The proposed approach aims to develop an automatic recognition system of fish species based on otolith shape analysis. From the 8-connected external contour of each otolith we extract the normal angles contour and then we represent it with Fourier coefficients. These coefficients are used to identify the classes of the otoliths using a neural network classification method. The approach was tested over 450 otolith images belonging to 15 species originated from a Moroccan Atlantic Ocean area. This database was collected and prepared in collaboration with the National Institute of Fisheries Research (INRH, Morocco). The experimental results showed a promising approach to classify otoliths.

A machine vision system is reported in this study for automatic recognition of the damages and symptoms on plant leaves from images. The system is based on a hybrid combination of three SVM classifiers including an individual classifier, which is used in parallel with a serial combination of two classifiers. The individual classifier adopts two types of features (texture and shape) to discriminate between the damages and symptoms. In serial architecture, the first classifier adopts the color features to classify the images; it considers the damages and/or symptoms that have a similar or nearest color belonging to the same class. Then, the second classifier is used to differentiate between the classes with similar color depending on the shape and texture features. A combination function is provided for comparing the decision of the individual classifier and of the serial architecture in order to achieve the final decision that represents the class of the form to be recognized. The tests of this study are carried out on six classes including three types of pest insects damages and three forms of fungal diseases symptoms. The results, with an overall recognition rate of 93.9 %, show the advantages of the proposed method compared to the other existing methods.

Plant taxonomy is a long-standing practice in botany. It uses the morphology of plant leaves to make categories. Leaf shape is one of the physical characteristics used to discriminate between plant species. This paper presents the characterisation of a leaf shape using the Convexity Measure of Polygons and the seven invariant moments in combination with other morphological features to improve leaf classification. The Convexity Measure of Polygons used in this paper is based on the minimum ratio obtained by dividing the rotated leaf-bounding perimeter of the associate bounding rectangle of the leaf shape. The proposed model is rotation, translation and scale invariant. It achieves a classification rate of 92 % on 400 leaves of 20 species, 99 % on 100 leaves of 4 species and 95 % on 1600 leaves of 32 species using a Multilayer Perceptron classifier. The proposed method out-performs several state-of-the-art methods when tested under the similar conditions, even with deformed leaves.

We present a novel and efficient room layout mapping strategy that does not reveal people’s identity. The system uses only a Kinect depth sensor instead of RGB cameras or a high-resolution depth sensor. The users’ facial details will neither be captured nor recognized by the system. The system recognizes and localizes 3D objects in an indoor environment, that includes the furniture and equipment, and generates a 2D map of room layout. Our system accomplishes layout mapping in three steps. First, it converts a depth image from the Kinect into a top-view image. Second, our system processes the top-view image by restoring the missing information from occlusion caused by moving people and random noise from Kinect depth sensor. Third, it recognizes and localizes different objects based on their shape and height for a given top-view image. We evaluated this system in two challenging real-world application scenarios: a laboratory room with four people present and a trauma room with up to 10 people during actual trauma resuscitations. The system achieved 80 % object recognition accuracy with 9.25 cm average layout mapping error for the laboratory furniture scenario and 82 % object recognition accuracy for the trauma resuscitation scenario during six actual trauma cases.

Quality control is an essential step in the textile manufacturing industry. There is a growing interest in the field of automation using computer vision for freeing human beings from the inspection task. In this paper, patterned fabric images are analyzed using entropy cues in order to detect different kinds of defects. In our proposal, we transform the test image to an entropy image in which the defects show low values and can be easily separated by a simple thresholding. Our method is evaluated and compared with previously proposed approaches, showing better results on an extensive database of real defective and non-defective fabrics.

In the molecular field, researchers analyze dynamics of polymers by microscopy: several measurements such as length and curvature are performed in their studies. To achieve correct analysis they need to extract the curve representing as good as possible the observed polymer shape which is a grayscale thick curve with noise and blur. We propose, in this paper, a method to extract such a curve. A polymer chain moves in a snake-like fashion (Reptation): it can self-intersect and form several complex geometries. To efficiently extract the different geometries, we generate the curve by computing a piecewise centerline browsing the shape by geodesics: each shape gives a set of separate geodesics. By fusion, we obtain the complete curve traveling the shape. To keep the correct curve orientation, the fusion is considered as a graph traversal problem. Promising results show that the extracted curve properly represents the shape and can be used for polymer study.

The chaotic cryptosystems have been widely investigated to provide fast and highly secure image encryption. In this paper, we introduce a novel cryptosystem for color image based on chaos by using a dynamic Look-Up Table (LUT). We utilized the Logistic Map chaotic system in order to benefit from its sensitivity to initial conditions.The result shows that the proposed cryptosystem have many characteristics such as high security, high sensitivity and high speed that can be applied in the encryption of color images. It is demonstrated that the NPCR = 99.6140 %, the UACI = 33.5448 % and entropy = 7.9984 can satisfy security and performance requirements. Simulations show that the proposed cryptosystem has high security and resist various typical attacks.

This paper aims to apply nonlinear estimation of chromophore concentrations: melanin, oxy-hemoglobin, deoxy-hemoglobin and shading to the real hyperspectral image of skin. Skin reflectance is captured in the wavelengths between 400 nm and 700 nm by hyperspectral scanner. Five-band wavelengths data are selected from skin reflectance. By using the cubic function which obtained by Monte Carlo simulation of light transport in multi-layered tissue, chromophore concentration is determined by minimizing residual sum of squares of reflectance.

One of the major issues related to dehazing methods (single or multiple image based) evaluation is the absence of the haze-free image (ground-truth). This is also a problem when it concerns the validation of Koschmieder model or its subsequent dehazing methods. To overcome this problem, we created a database called CHIC (Color Hazy Image for Comparison), consisting of two scenes in controlled environment. In addition to the haze-free image, we provide 9 images of different fog densities. Moreover, for each scene, we provide a number of parameters such as local scene depth, distance from the camera of known objects such as Macbeth Color Checkers, their radiance, and the haze level through transmittance. All of these features allow the possibility to evaluate and compare between dehazing methods by using full-reference image quality metrics regarding the haze-free image, and also to evaluate the accuracy of the Koschmieder hazy image formation model.

We propose a method of hyperspectral unmixing for the linear mixing model (LMM) while both the spectral signatures of endmembers and their fractional abundances are unknown. The proposed algorithm employs the non-negative matrix factorization (NMF) method as well as simultaneous (collaborative) sparse regression model. We formulate the NMF problem along with an averaging over the $$\ell _2$$-norm of the fractional abundances so-called $$\ell _{2,q}$$-norm term. We show that this problem can be efficiently solved by using the Karush-Kuhn-Tucker (KKT) conditions. Our simulations show that the proposed algorithm outperforms the state-of-the-art methods in terms of spectral angle distance (SAD) and abundance angle distance (AAD).

Text detection in natural scene image is still open research topics. Particularly, for Arabic text, a very few studies have been proposed. In this paper, we propose a method for Arabic text detection in natural scene image based on the color homogeneity. Starting from the MSER idea and instead of relying on a range of unique thresholds we calculate a range of pairs of thresholds for each channel in the RGB space in order to generate a set of binary maps. Following extraction of connected components of each binary map we apply a first filtering according to a stability criterion of the written texts to extract candidate components regardless of the language. Then, through the characteristics of the Arabic script we make a second screening to found candidates to keep only those that define a text in the Arabic language.

This paper presents a method to measure spectral reflectance and 3D shape of an object. For realizing these measurements, we applied a multi-primary image projector as a computational illumination system. This multi-primary image projector employs a light source which is programmable and can reproduce any spectral power distributions. In other words, the projector can reproduce 2D pattern projections with arbitrary spectra. In our actual measurements, we developed an imaging system by synchronizing the multi-primary image projector and a highspeed monochrome camera. First, the surface spectral reflectance of an object in a darkroom was obtained based on a finite-dimensional linear model of spectral reflectances. In the spectral reflectance measurements, nine basis images were projected and captured by the synchronized imaging system. Then spectral reflectance at each camera image coordinate was estimated from the captured nine images. Next, structured lights were projected for reconstructing 3D shape. We applied eight binary image projections and a conventional 3D shape reconstruction algorithm to our study. In summary, seventeen images were projected and captured for measuring spectral reflectance and 3D shape. The projection and capturing speed of the seventeen images is 0.085 s on the system specification. In the validation experiments, we could obtain spectral reflectance of X-rite ColorChecker with the average color difference $$\varDelta E_{ab}^{*}$$ of approximately 4. We also confirmed that precise 3D shapes could be reconstructed by our method.

In this paper, we propose a fast solution for the problem of illuminant color estimation. We present a physics-based algorithm that uses the mean projections maximization assumption. We investigated this hypothesis on a large images dataset and used it afterwords to estimate the illuminant color. The proposed algorithm reduces the illuminant estimation problem to an uncentred PCA problem. The evaluation of the algorithm on two well-known image datasets results in lower angular errors.

Multispectral images have been studied in various fields such as remote sensing and sugar content prediction in fruits. One of the systems that captures multispectral images uses a multispectral filter array based on a color filter array. In this system, demosaicking processing is required because the captured multispectral images are mosaicked. However, demosaicking is more difficult for multispectral images than for RGB images owing to the low density between the observed pixels in multispectral images. Therefore, we propose a demosaicking method for multispectral images based on spatial gradient and inter-channel correlation. Experimental results demonstrate that our proposed method outperforms the existing methods and is effective.

This paper presents a new method of generating a high-resolution image from a low-resolution image. We use a sparse representation based model for low-resolution image patches. We use large patches instead of small ones of existing methods. The size of the dictionary must be large to guarantee its completeness. For each patch in the low-resolution image, we search for similar patches in the dictionary to obtain a sub-dictionary. To define the similarity and to speed up the searching process, we present a Restricted Boltzmann Machine (RBM) based binary encoding method to get binary codes for the low-resolution patches, and use Hamming distance to describe the similarity. With the KNN dictionary of each low-resolution patch, we use a sparse representation method to get its high-resolution version. Experimental results illustrate that our method outperforms other methods.

Super-resolved image enhancement is of great importance in medical imaging. Conventional methods often require multiple low resolution (LR) images from different views of the same object or learning from large amount of training datasets to achieve success. However, in real clinical environments, these prerequisites are rarely fulfilled. In this paper, we present a self-learning based method to perform super-resolution (SR) from a single LR input. The mappings between the given LR image and its downsampled versions are modeled using support vector regression on features extracted from sparse coded dictionaries, coupled with dual-tree complex wavelet transform based denoising. We demonstrate the efficacy of our method in application of cardiac MRI enhancement. Both quantitative and qualitative results show that our SR method is able to preserve fine textural details that can be corrupted by noise, and therefore can maintain crucial diagnostic information.

In this paper, speaker classification using supervised hierarchical clustering is provided. Bounded generalized Gaussian mixture model with ICA is adapted for statistical learning in the clustering framework. In the presented framework ICA mixture model is learned through training data and the posterior probability is used to split the training data into clusters. The class label of the training data is further selected to mark each cluster into a specific class. The cluster-class information from the training process is taken as reference for the classification of test data into different speaker classes. This framework is employed for the gender and 10 speakers classification and TIMIT and TSP speech corpora are selected to validate and test the classification framework. This classification framework also validate the statistical learning of our recently proposed ICA mixture model. In order to examine the performance of the ICA mixture model, the classification results are compared with same framework using Gaussian mixture model. It is observed that: (i) presented clustering framework performs well for the speaker classification, (ii) ICA mixture model outperforms Gaussian mixture model in the statistical learning based on the classification accuracy for gender and multi-class scenarios.

Automatic Speaker Discrimination consists in checking whether two speech signals belong to the same speaker or not. It is often difficult to decide what could be the best classifier to use in some specific circumstances. That is why, we implemented nine different classifiers, namely: Linear Discriminant Analysis, Adaboost, Support Vector Machines, Multi-Layer Perceptron, Linear Regression, Generalized Linear Model, Self Organizing Map, Second Order Statistical Measures and Gaussian Mixture Models. Moreover, a special feature reduction was proposed, which we called Relativistic Speaker Characteristic (RSC). On the other hand we further intensified the feature reduction by adding a second step of feature transformation using a Principal Component Analysis (PCA). Experiments of speaker discrimination are conducted on Hub4 Broadcast-News. Results show that the best classifier is the SVM and that the proposed feature reduction association (RSC-PCA) is extremely efficient in automatic speaker discrimination.

Speaker discrimination consists in checking whether two (or more) speech segments belong to the same speaker or not. In this framework, we propose a new approach developed for the task of speaker discrimination, this approach results from the fusion between a neural network classifier (NN) and a statistical classifier, this fusion is obtained once by combining the scores of the simple classifiers weighted by some confidence coefficients and another time, by using the scores of the statistical classifier as an additional input of the Multi-Layer Perceptron (MLP), in order to optimize the NN training (Hybrid model).In one hand, we notice that the fusion has improved the results obtained by each approach alone and in the other hand we notice that the fusion using the sum of weighted scores, obtained by each classifier alone, seems to be better than the hybrid method. The experiments, done on a subset of Hub4 Broadcast News database, have shown the efficiency of that fusion in speaker discrimination, where the Equal Error Rate (EER) is about 7 %, with short segments of 4 s only.

Frog call classification has received increasing attention due to its importance for ecosystem. Traditionally, the classification of frog calls is solved by means of the single-instance single-label classification classifier. However, since different frog species tend to call simultaneously, classifying frog calls becomes a multiple-instance multiple-label learning problem. In this paper, we propose a novel method for the classification of frog species using multiple-instance multiple-label (MIML) classifiers. To be specific, continuous recordings are first segmented into audio clips (10 s). For each audio clip, acoustic event detection is used to segment frog syllables. Then, three feature sets are extracted from each syllable: mask descriptor, profile statistics, and the combination of mask descriptor and profile statistics. Next, a bag generator is applied to those extracted features. Finally, three MIML classifiers, MIML-SVM, MIML-RBF, and MIML-kNN, are employed for tagging each audio clip with different frog species. Experimental results show that our proposed method can achieve high accuracy (81.8 % true positive/negatives) for frog call classification.

In this paper, we propose an adaptive frequency scale filter bank to perform frog call classification. After preprocessing, the acoustic signal is segmented into individual syllables from which spectral peak track is extracted. Then, syllable features including track duration, dominant frequency, and oscillation rate are calculated. Next, a k-means clustering technique is applied to the dominant frequency of syllables for all frog species, whose centroids are used to construct a frequency scale. Furthermore, one novel feature named bandpass filter bank cepstral coefficients is extracted by applying a bandpass filter bank to the spectral of each syllable, where the filter bank is designed based on the generated frequency scale. Finally, a k-nearest neighbour classifier is adopted to classify frog calls based on extracted features. The experiment results show that our proposed feature can achieve an average classification accuracy of 94.3 % which outperforms Mel-frequency cepstral coefficients features (81.4 %) and syllable features (88.1 %).

Accurate classification of organisms into taxonomical hierarchies based on genomic sequences is currently an open challenge, because majority of the traditional techniques have been found wanting. In this study, we employed mitochondrial DNA (mtDNA) genomic sequences and Digital Signal Processing (DSP) for accurate classification of Eukaryotic organisms. The mtDNA sequences of the selected organisms were first encoded using three popular genomic numerical representation methods in the literature, which are Atomic Number (AN), Molecular Mass (MM) and Electron-Ion Interaction Pseudopotential (EIIP). The numerically encoded sequences were further processed with a DSP based cepstral analysis to obtain three sets of Genomic Cepstral Coefficients (GCC), which serve as the genomic descriptors in this study. The three genomic descriptors are named AN-GCC, MM-GCC and EIIP-GCC. The experimental results using the genomic descriptors, backpropagation and radial basis function neural networks gave better classification accuracies than a comparable descriptor in the literature. The results further show that the accuracy of the proposed genomic descriptors in this study are not dependent on the numerical encoding methods.

In this paper, we address the problem of the diffusion signal reconstruction from a limited number of samples. The HARDI (High Angular Resolution Diffusion Imaging) technique was proposed as an alternative to resolve the problems of crossing fibers in the case of Diffusion Tensor Imaging (DTI). However, it requires a long scanning time for the acquisition of the Diffusion Weighted (DW) images. This fact makes hard the clinical applications. We propose here a novel geometrical approach to accurately estimate the HARDI signal from a few number of DW images. The missing diffusion data are obtained according to their neighborhood from a reduced set of diffusion directions on the sphere of the q-space. The experimentations are performed on both synthetic data and many digital phantoms simulating crossing fibers on the brain tissues. The obtained results show the accuracy of the reconstruction of the Fiber Orientation Distribution (FOD) function from the estimated diffusion signal.

Objectives: Clinical intervention for non-breathing newborns due to birth asphyxia needs to be conducted within the first minute of life. The responses of the babies are affected by complicated interactions between physiological conditions of the newborns and the combination of various clinical treatments, e.g., drying thoroughly, stimulation, manual bag-mask ventilation, chest compression, etc. Previously, we have proposed methods to detect and parameterize various events regarding bag mask ventilation. However, the outcome of the resuscitation is likely influenced by not only ventilation but also other therapeutics activities. The detection of the existence of activities using information from acceleration signals is illustrated in this paper. Methods: Short time energy of the acceleration signal is calculated. A thresholding method is applied on the amplitude of the energy signal to determine activity or rest. Results: The average sensitivity and specificity of the detection of activities are 90 % and 80 % respectively. Conclusions: The performance of the detection algorithm indicates the possibility to use acceleration signal to detect the presence of various activities during resuscitation procedure.

In this work, we propose a bootstrapped generalized mixture estimation algorithm for synthetic aperture radar image segmentation. The Bootstrap sampling reduces the dependence effect of pixels in real images, and reduces segmentation time. Given an original image, we randomly select small representative set of pixels. Then, a generalized expectation maximization algorithm based on optimal Bootstrap sample is released for mixture identification. The generalized aspect comes from the use of distributions from the Pearson system. We validate the proposed algorithm on the classification of SAR images. The results we obtain show that the bootstrap sampling method yield the same accuracy and robustness of image classification as the basic algorithm while reducing time computing. This fact make possible the integration of such technique in real time applications.

Palm groves are one of the most characteristic agro-ecosystems of Morocco. Therefore, conservation and monitoring have become a primary objective, not just from an environmental and landscaping point of view but also from the socio-economic. In this context, remote sensing presents an effective tool to map palm groves, to count palm trees and to detect their possible diseases.The present study attempts to map palm trees from very high resolution WorldView 2 (WV 2) imagery, using a new supervised contextual classification method based on Markov Random Fields and palm trees shadow orientation. A combined layer of pan-sharpened multi-spectral (MS) bands and eight mean texture measures based Gray Level Co-occurrence Matrices (GLCM) were used as input variables. Total accuracy of 83.4 % palm trees detection was achieved. Using a decision criterion based on palm trees: shape, shadow orientation and the distance, the total accuracy of palm trees detection reached 88.1 %.

Unmanned landing on unknown terrain such as planetary surfaces requires the in-situ estimation of surface irregularities like craters, ridges and other deformities. Moreover, to facilitate safe landing, the surface estimation has to be done in as little time as possible. In this paper, we present an algorithm to address the above two issues in the context of crater presence on the terrain. Detection of craters is done on images of the probable landing surfaces and the computation time required for the detection is subsequently reduced in the proposed method using image analysis approaches like standard deviation filtering, morphological operations and validation of crater presence by texture extraction. We have achieved a 85–89% true positive (TP) rate on large craters and 79–82 % TP rate on small craters. We have conducted our experiments on real images of Mars and the Moon, collected by space-crafts named 2001 Mars Odyssey and the Lunar Reconnaissance Orbiter, respectively. Empirical evidences indicate that the proposed method achieves a commendable TP rate and a subsequent improvement in the time required for detection as compared to existing methodologies.

Oil reserves are one of the core interests of a country. The detection of oil tanks is a very important task. So far, most studies only focus on the detection task itself. But the strategic value of different types of oil tanks is obviously different. So we furtherly divide oil tanks into two types: flat crest and cone-shaped crest. In this paper, a four-step method is adopted: (1) prepare dataset; (2) train the classifier; (3) extract candidate regions and (4) classification. The deep network (CNN) Krizhevsky used on cifar-10 dataset is used to train the classifier and ELSD is used to extract candidate regions. In addition, some clustering tricks are used to determine an only candidate region to solve the double-detection problem. The experimental results show that this method can detect and distinguish different types of oil tanks with outstanding performance.

The goal of an image watermarking scheme is to embed a watermark that is robust against various types of attacks while preserving the perceptual quality of the cover image. In this paper, a discrete cosine transform and singular value decomposition based digital image watermarking scheme that makes use of Arnold transform is proposed. The basic idea behind the proposed Arnold transform based watermarking scheme is to improve the robustness of the watermarked image, while providing complete security to the embedded watermark. The new scheme is shown to retain the perceptibility of the cover image in the watermarked image due to the discrete cosine transform and singular value decomposition based watermark embedding. Extensive experiments are performed to demonstrate the performance of the proposed scheme in providing security to the watermark content, preserving the perceptibility of the cover image and in being robust against various types of attacks on the watermarked image.

Modeling the human visual system has become an important issue in image processing such as compression, evaluation of image quality and digital watermarking. In this paper we present a spatial JND (Just Noticeable-Difference-) model that uses a texture selector based on Faber-Schauder wavelets lifting scheme. This texture selector identify non-uniform and uniform areas. That allows to choose between JNDs models developed by Chou and Qi. The chosen JND will determine the value of the embedding strength in each pixel, related to the identified region. Results show that by this process, we can generally ameliorate the visual quality with the same robustness.

The modification of digital content becomes easier owing to the technology development. The need of authenticating digital content is increasing. Different fragile watermarking methods have been proposed for image authentication. In some applications low complexity algorithm is required, such as real-time and video processing applications. We propose in this paper, a fragile watermarking scheme for image authentication using Faber Schauder Discrete Wavelet Transform (FSDWT) and Singular Value decomposition (SVD) where the data to be embedded is a logo. The Watermark was generated by applying XOR operation between the logo’s bits and the bits of dominant blocks singular values of the image to authenticate. These dominant blocks are obtained by applying FSDWT to that image. Any image modification will result in significant change of the dominant blocks singular values, which helps the watermarking scheme to detect the authenticity of the image. Furthermore, FSDWT is composed of simple operations, hence, the algorithm has a low complexity.

We present a novel and very efficient software architecture designed for JPEG 2000 coders. The proposed method employs a strip-based data processing technique while performing a single-pass multi-scale wavelet transform. The overall compression chain is driven by incoming data while the fragments of the resulting bitstream are produced immediately after loading the corresponding data and additionally in parallel. The method is friendly to the CPU cache and nicely exploits the SIMD capabilities of the modern CPUs. Implanted into reference OpenJPEG implementation, our method has significantly better performance in terms of the execution time.

Perspective deformation is one of the major issues in print-cam attacks for image watermarking. In this paper we adapt to print-cam process a Fourier watermarking method developed by our team, for print-scan attacks. Our aim is to resist to perspective distortions of print-cam image watermarking for ID images for industrial application. A first step consists of geometrical correction of the perspective distortions, then Fourier based watermarking is used. Experimental results of the improved method in Fourier domain show better robustness compared to existing print-cam methods.

3D meshes are subject to various visual distortions during their transmission and geometrical processing. Several works have tried to evaluate the visual quality using either full reference or reduced reference approaches. However, these approaches require the presence of the reference mesh which is not available in such practical situations. In this paper, the main contribution lies in the design of a computational method to automatically predict the perceived mesh quality without reference and without knowing beforehand the distortion type. Following the no-reference (NR) quality assessment principle, the proposed method focuses only on the distorted mesh. Specifically, the dihedral angles are firstly computed as a surface roughness indexes and so a structural information descriptors. Then, a visual masking modulation is applied to this angles according to the main characteristics of the human visual system. The well known statistical Gamma model is used to fit the dihedral angles distribution. Finally, the estimated parameters of the model are learned to the support vector regression (SVR) in order to predict the quality score. Experimental results demonstrate the highly competitive performance of the proposed no-reference method relative to the most influential methods for mesh quality assessment.

Due to measurement errors or interference noise, Kinect depth maps exhibit severe defects of holes and noise, which significantly affect their applicability to stereo visions. Filtering and inpainting techniques have been extensively applied to hole filling. However, they either fail to fill in large holes or introduce other artifacts near depth discontinuities, such as blurring, jagging, and ringing. The emerging reconstruction-based methods employ underlying regularized representation models to obtain relatively accurate combination coefficients, leading to improved depth recovery results. Motivated by sparse representation, this paper advocates a similarity and position constrained sparse representation for Kinect depth recovery, which considers the constraints of intensity similarity and spatial distance between reference patches and target one on sparsity penalty term, as well as position constraint of centroid pixel in the target patch on data-fidelity term. Various experimental results on real-world Kinect maps and public datasets show that the proposed method outperforms state-of-the-art methods in filling effects of both flat and discontinuous regions.

Digital documentation of cultural heritage requires high quality spatial and color information. However the 3D data accuracy is already sufficiently high for many applications, the color representation of surface remains unsatisfactory. In this paper we describe issues encountered during 3D and color digitization based on a real-world case study. We focus on documentation of the King’s Chinese Cabinet at Wilanów Palace (Warsaw, Poland). We show the scale of the undertaking and enumerate problems related to high resolution 3D scanning and reconstruction of the surface appearance despite object gloss, uneven illumination, limited field of view and utilization of multiple light sources and detectors. Our findings prove the complexity of cultural heritage digitization, justify the individual approach in each case and provide valuable guidelines for future applications.

Recovering detailed morphological information from archaeological or paleontological material requires extensive hands-on time. Creating 3D scans based on e.g. computed tomography (CT) will recover the geometry of the specimen, but can inflict bimolecular degradation. Instead, we propose a fast, inoffensive and inexpensive 3D scanning modality based on structured light, suitable for capturing the morphology and the appearance of specimens. Benefits of having 3D models are manifold. The 3D models are easy to share among researchers and can be made available to the general public. Advanced morphological modelling is possible with accurate description of the specimens provided by the models. Furthermore, performing studies on models reduces the risk of damage to the original specimen. In our work we employ a high resolution structured light scanner for digitalizing a collection of 8500 year old human skulls from Brazil. To evaluate the precision of our set-up we compare the structured light scan to micro-CT and achieve sub-millimetre difference. We analyse morphological features of the Brazilian skulls using manual landmarks, but a research goal is to automate this, fully utilize the dense 3D scans, and apply the method to many more samples.