The emergence of large-scale digital image collections in industrial Digital Asset Management (DAM) systems presents significant challenges, requiring classification frameworks capable of managing hierarchical taxonomies and multi-label categorization tasks. This paper presents a comprehensive framework for multi-label hierarchical image classification that bridges the gap between academic research and industrial applications. We propose: (1) a modular architecture with generic hierarchical heads adaptable to any visual backbone, featuring a conditional concatenation strategy for taxonomic consistency; (2) a specialized training approach combining selective backbone fine-tuning with a composite loss function that simultaneously optimizes three metrics. (3) extensive validation on both academic benchmarks and real-world industrial datasets. Experimental results demonstrate competitive performance on academic datasets while achieving promising results on industrial multi-hierarchy scenarios.

In the pneumatic industry, quality control is an essential step in assessing tire compliance. Artificial neural networks are increasingly used to accomplish this task. Their training requires a large number of images of the controlled products. However, at the launch production of a new tire, the lack of images causes a performance loss for the network. To solve this problem, we propose to translate perfect tires computer-based images into ad hoc manufacturing context-realistic ones as pre-manufacturing step to improve robustness and ensure production quality. The challenging work is to extract features in real images and apply them to computer-based images while maintaining the original geometry. In the paper, we propose Prefab-GEN, a novel architecture based on Cycle-GAN. In the generator part, an Inception U-Net architecture is developed to enforce geometrical structure conversion and extract more detailed features. The qualitative and quantitative evaluation on tire dataset shows improvements compared with state-of-art.

This paper presents a different way to improve the resistance of digital watermarking. Using the well known Lattice QIM in the spatial domain, we analyze the interest of using a different kind of error correcting codes: rank metric codes. These codes are already used in communications for network coding but not used in the context of watermarking. In this article, we show how this metric permits to correct errors with a specific structure and is adapted to specific image attacks. We propose a first study to validate the concept of rank metric for watermarking process. For this, we use these codes to obtain invariance against luminance additive constant change.

We propose a nonrigid registration method whose motion estimation is cast into a feature matching problem under the Log-Demons framework using Graph Wavelets. We investigate the Spectral Graph Wavelets (SGWs) to capture the shape features of the images. The SGWs are more adapted to learn the spatial and geometric organization of data with complex structures than the classical wavelets. Our experiments on T1 brain images and endomicroscopic images show that this method outperforms the existing nonrigid image registration techniques (i.e. Log-Demons and Spectral Log-Demons) with improved similarity values.

Can we build a feature-based analysis that fully characterizes images? The literature answers with edge-based reconstruction methods inspired by Marr's paradigm but limited to the greyscale case. This paper studies the color case. A new sparse representation is carried out with the monogenic concept and the Mallat-Zhong wavelet maxima method. Our monogenic maxima provide efficient contour shape and color characterization, as a sparse set of local features including amplitude, phase, orientation and ellipse parameters. This rich description takes the wavelet maxima representation further towards the wide topic of keypoint analysis. We propose a reconstruction process that retrieves the image from its monogenic maxima. While known works all rely on constrained optimization, implying an iterative use of the filterbank, we propose to interpolate the data in the feature domain by exploiting the visual knowledge from the feature-set. This direct retrieval is accurate enough so that no iteration is required. The main question is finally answered with comparative experiments. It is shown that a reasonably small amount of features is sufficiently informative for visually appealing image retrieval. The features appear numerically stable to rotation, and can be intuitively simplified to perform image regularization.

Over the last few years, considering 3D vectors as one color information instead of three independent vector components has significantly improved the color watermarking field. There have been some research about perceptual approaches but there are few about the perception of color differences of the human vision system (HVS) for watermarking applications. This paper propose a new color watermarking algorithm able to minimize the perception of color differences. It understands color information as the HVS does. It can easily be adapted to many watermarking schemes such as quantization based schemes [1] or spread spectrum insertion techniques. This algorithm is based on a psychovisual model of the human eye studied by D. Alleysson [2]. The results showed good improvements in terms of watermark invisibility and robustness to image processings: we compared quantization methods working in grayscale and its color adaptation to show model stability or improvement.

Blind steganalysis techniques are able to detect the presence of secret messages embedded in digital media files, such as images, video, and audio, with an unknown steganography algorithm. This paper present an image steganalysis method based on Evidential K-Nearest Neighbors (EV-knn). Originality of this work is the use of theoretical framework of Belief functions on different subspaces of features vectors. Classifications obtained in subspaces are combined using specific combination function and to provide classification of a given image (cover or stego). The proposed approach is evaluated with the classical nsf5 steganographic method that hides messages in JPEG images. Compared to Ensemble Classifier steganalysis algorithm, the proposed approach significantly increases the performance of classification.

Splicing operation is a common technique used for image forgery. In this paper, we propose a new method for detecting the image splicing by revealing inconsistencies of the illuminant color in the object regions. For this, we first divide a given image into some horizontal and vertical bands. After that, the illuminant of each band is estimated with the generalized grey-world algorithms. For each illuminant estimation algorithm, the fake patches produced by computing the intersection between the forged horizontal and vertical bands are represented in a forgery detection map. Finally the spliced regions are shown by combining all splicing detection maps. Our proposed method is validated with several test cases. Additionally, we compare the performance of our method with two existing methods on the dataset CASIA V2.0. The results show that the proposed method works well with minimal human interaction.

We define a new color extension for the monogenic representation of images by using an elliptical tri-valued oscillation model jointly with the vector structure tensor formalism. The proposed method provides a rich local colorimetric and geometric analysis, in particular a color phase concept, which can be computed by a numerically stable algorithm. This representation is finally used to estimate the local frequency of color images.

In this paper, we propose a new polynomial based texture representation method for extracting information about facial expressions. While many appearance-based methods have been proposed over the years to improve the performance of facial expression recognition, most descriptors are usually unable to both provide precise multi-scale / multi-orientation analysis and handle the redundancy problem effectively. We will explain how coefficients obtained from polynomial projections of pixel intensities on a complete basis can be used for compact, hierarchical image approximation and structural analysis. We have tested our approach on two publicly available databases and achieved encouraging results comparable to the state of the art.

The estimation of dependence relationships between variables is generally performed using probabilistic models. However, these models are not adapted to imprecise data and they cannot easily take into account symbolic information such as experts opinions. On the contrary, evidence theory also called theory of belief function, allow to integrate these kinds of uncertainties. In this paper we propose regression analysis based on a fuzzy extension of belief function theory, applied to image quality assessment problem. For a given input vector x of relevant images feature, the method provides a prediction regarding the value of the output variable y which represents the score of subjective image quality test, namely the DMOS value. To validate the proposed approach, experiments are conducted on LIVE image database. The proposed measure is compared with algorithms based on general regression as neural networks and Support Vector Machine (SVM). The framework of this paper is of nature subjective and results show that our approach performs well and illustrate the interest of the theory of belief function in this context.

This paper proposes a new reduced-reference metric based on the modeling of Quaternionic Wavelet Transform (QWT) coefficients from Information Criteria (IC). To obtain the reduced-references, we will model the QWT coefficients using probability density functions (pdf) whose parameters are used as reduced-references. IC are proposed in order to build the optimal histograms of the QWT coefficients to get most likely pdf of these. In the mixture model, IC are also used to obtain the number of distribution. From these models, we propose a measure of degradation by comparing probability density functions of the reference image and the distributions of the degraded image of the QWT subbands. We shall demonstrate that one phase of the QWT provides relevant information in the Image Quality Assessment. Tests confirmed the potentiality of this information and showed that the QWT produces a better coefficient of correlation with the Human Visual System than the Discrete Wavelet Transform.

We introduce a new constrained minimization problem that performs template and pattern detection on a multispectral image in a compressive sensing context. We use an original minimization problem from Guo and Osher that uses L1minimization techniques to perform template detection in a multispectral image. We first adapt this minimization problem to work with compressive sensing data. Then, we extend it to perform pattern detection using a formal transform called the specialization along a pattern. That extension brings out the problem of measurement reconstruction. We introduce shifted measurements that allow us to reconstruct all measurement with a small overhead, and we give an optimality constraint for simple patterns. We present numerical results showing the performances of the original minimization problem and the compressed ones with different measurement rates and applied on remotely sensed data.

Methods based on partial differential equations (PDE) become increasingly one of the methods of image processing. Recently a diffusion method is appeared, it allows to generalize the diffusion to the complex domain by the injection of a complex number in the heat equation. For small phase angles, the linear process generates the Gaussian and Laplacian pyramids (scale-spaces) simultaneously, depicted in the real and imaginary parts, respectively. The imaginary value serves as a robust edge-detector with increasing confidence in time, thus handles noise well and may serve as a controller for nonlinear processes. In this article we propose to extend this concept by introducing a notion of directionality in such a way as each equation of the system will correspond to a specific direction. It is in our interests to use higher order algebra to adapt the process to the four discrete directions. Then we will focus on the imaginary parts for developing a nonlinear scheme.

This paper adapts a new template matching and target detection algorithm in multispectral images to a compressive sensing strategy. That template matching algorithm found in [1] relies on particular properties of L1 minimization algorithms to succeed. We propose a new algorithm that is reconstructing in a single step the location of a given signature of interest bypassing the image reconstruction and the template matching algorithm on that image. For that purpose, we use a modified split Bregman algorithm with various regularizers. We conduct numerical experiments on real-world multispectral image.

Abstract This article details the conception, design, development and analysis of invisible, blind and robust color image watermarking algorithms based on the wavelet transform. Using error correcting codes, the watermarking algorithms are designed to be robust against intentional or unintentional attacks such as JPEG compression, additive white Gaussian noise, low pass filter and color attacks (hue, saturation and brightness modifications). Considering the watermarking channel characterized by these attacks, repetition, Hamming, Bose Chaudhuri Hocquenghem and Reed-Solomon codes are used in order to improve the robustness using different modes and appropriate decoding algorithms. The article compares the efficiency of different type of codes against different type of attacks. To the best of our knowledge this is the first time that the effect of error-correcting codes against different attacks are detailed in a watermarking context in such a precise way: describing and comparing the effect of different classes of codes against different type of attacks. This article clearly shows that list decoding of Reed-Solomon codes using the algorithm of Sudan exhibits good performance against hue and saturation attacks. The use of error correcting codes in a concatenation mode allows the non-binary block codes to show good performance against JPEG compression, noise and brightness attacks.

Monogenic wavelets offer a geometric representation of grayscale images through an AM-FM model allowing invariance of coefficients to translations and rotations. The underlying concept of local phase includes a fine contour analysis into a coherent unified framework. Starting from a link with structure tensors, we propose a nontrivial extension of the monogenic framework to vector-valued signals to carry out a nonmarginal color monogenic wavelet transform. We also give a practical study of this new wavelet transform in the contexts of sparse representations and invariant analysis, which helps to understand the physical interpretation of coefficients and validates the interest of our theoretical construction.

We propose a new color extension of the monogenic wavelet transform. Monogenic wavelets give a coherent representation of scalar images through a local phase concept and an underlying orientation analysis. We here define a color extension of the monogenic framework. The underlying local geometric analysis and phase concept are generalized by using the color structure tensor. Resulting transform appears to be a clear improvement of our previous work - that is the only proposition of color monogenic wavelets to our knowledge. This is shift and rotation invariant and efficiently represents multiscale color lines and edges.

We define a color monogenic wavelet transform. This is based on the recent grayscale monogenic wavelet transform and a non-marginal extension to color signals. To our knowledge, wavelet based color image processing schemes have always been made by using a grayscale tool separately on color channels. This may have some unexpected effect on colors because those marginal schemes are not necessarily justified. Here we propose a definition that considers a color (vector) image right at the beginning of the mathematical definition and so brings an actual color wavelet transform - which has not been done so far to our knowledge. This so provides a promising multiresolution color geometric analysis of images.

This paper proposes a new texture classifier based on the Quaternionic Wavelet Transform (QWT). This recent transform separates the informations contained in the image better than a classical wavelet transform (DWT), and provides a multiscale image analysis which coefficients are 2D analytic, with one near-shift invariant magnitude and a phase, that is made of three angles. The interpretation and use of the QWT coefficients, especially the phase, are discussed, and we present a texture classifier using both the QWT magnitude and the QWT phase of images. Our classifier performs a better recognition rate than a standard wavelet based classifier.

This paper presents a perception based watermarking algorithm for perceptually uniform color spaces. The image is watermarked using a blind watermarking scheme in the contourlet domain for the RGB, CIELAB, Y UV and AC1C2color spaces. The contourlet transform is used to decompose the image into directional subbands at multiple levels for all the color spaces. The contourlet transform allows to models the spatial frequency selectivity of the human visual system. The invisibility results of all the color spaces are compared with each other using CIEDE2000 and SSIM metrics. The visual distortion for each color component and its interaction with other color components is considered for all the subbands. Subjective tests are carried out to further validate the objective results. The results of the testing procedures show that the YUV color space is best suited for watermark insertion in terms of human perception.

In this paper, a method of specific colored area extraction into a color image is presented. Usual color segmentation or edge detection operators perform a global processing on the image. Thus, every area into an image is detected. The whole image is subdivided into several region by labeled pixels, or boundaries in case of an edge detection. Our purpose is to extract only specific areas into an image sharing a specific color attribute, hence sharing a common color based on a specific hue. The quaternionic geometrical transformation into RGB color space is used. A specific hue axis is defined. From this axis, a criterion defines a color subspace into an RGB color space using the quaternionic HSI interpretation of RGB vectors.

In this paper, we present the use of differential geometry for the segmentation of multispectral images, which allows us to unify several known methods including projecting onto a particular axis or a particular plan. This is done by choosing a metric tensor on the feature space computing the pullback of the metric tensor and applying standard Di Zenzo algorithm.

In this paper, an underwater image enhancement method using quaternions is presented. This work aims to improve color rendition and contrast of the objects, as if the scene has been taken out of water. This method is based on light attenuation inversion after processing a color space contraction using quaternions. Applied to the white, the attenuation gives a hue vector characterizing the water color. Using this reference axis, geometrical transformations into the color space are computed with quaternions. Pixels of water areas of processed images are moved to gray or colors with a low saturation whereas the objects remain fully colored. Thus, the contrast of the observed scene is significantly improved and the difference between the background and the rest of the image is increased, giving a first approach towards a pre-segmentation step.

In this paper we propose the use of list decoding of Reed Solomon codes to improve the robustness of blind frequency domain watermarking schemes in the presence of attacks, specially aimed at the colour aspect of the schemes. List decoding allows to decode errors well beyond the normally used bounded distance decoding algorithms and shows significant improvement when the code rates are low. We compare the results of different families of error correcting codes in order to improve the robustness of a blind, wavelet transform based colour image watermarking scheme. We use four families of error correcting codes to improve the robustness of the watermarking scheme (repetition, Hamming, BCH, and Reed Solomon). The experimental results show consistency with the theoretical approximations and the watermarked images with lower code rates are resistant to a variety of attacks.

It is well known that removing cloud-contaminated portions of a remotely sensed image and then filling in the missing data represent an important photo editing cumbersome task. In this paper, an efficient inpainting technique for the reconstruction of areas obscured by clouds or cloud shadows in remotely sensed images is presented. This technique is based on the Bandelet transform and the multiscale geometrical grouping. It consists of two steps. In the first step, the curves of geometric flow of different zones of the image are determined by using the Bandelet transform with multiscale grouping. This step allows an efficient representation of the multiscale geometry of the image's structures. Having well represented this geometry, the information inside the cloud-contaminated zone is synthesized by propagating the geometrical flow curves inside that zone. This step is accomplished by minimizing a functional whose role is to reconstruct the missing or cloud contaminated zone independently of the size and topology of the inpainting domain. The proposed technique is illustrated with some examples on processing aerial images. The obtained results are compared with those obtained by other clouds removal techniques.

A framework for active contour segmentation in vector-valued images is presented. It is known that the standard active contour is a powerful segmentation method, yet it is susceptible to weak edges and image noise. The proposed scheme uses foveal wavelets for an accurate detection of the edges singularities of the image. The foveal wavelets introduced by Mallat (2000) are known by their high capability to precisely characterize the holder regularity of singularities. Therefore, image contours are accurately localized and are well discriminated from noise. Foveal wavelet coefficients are updated using the gradient descent algorithm to guide the snake deformation to the true boundaries of the objects being segmented. Thus, the curve flow corresponding to the proposed active contour holds formal existence, uniqueness, stability and correctness results in spite of the presence of noise where traditional snake approach may fail.

Visual tasks often require a hierarchical representation of images in scales ranging from coarse to fine. A variety of linear and nonlinear smoothing techniques, such as Gaussian smoothing, anisotropic diffusion, regularization, wavelet thresholding etc... have been proposed. In this work, we propose a geometrical multiscale anisotropic diffusion based on the geometrical flow for denoising multivalued images. The geometrical flow is determined by the Bandelet transform of the image being processed. Consequently, the image is segmented into a quadtree where each square regroups pixels sharing the same geometrical flow direction. The motivation of this work is to introduce a new multiscale multistructure bandelet-based diffusion tensor to adjust the anisotropic diffusion toward the direction of the optimal geometrical flow. Therefore, multiple dyadic squares in the quadtree have multiple structure tensors. Hence, a more accurate geometrically driven noise suppression is obtained where the homogeneity of different image regions is well maintained.

In this paper, we propose an implementation of the 3-D Ridgelet transform: the 3-D discrete analytical Ridgelet transform (3-D DART). This transform uses the Fourier strategy for the computation of the associated 3-D discrete Radon transform. The innovative step is the definition of a discrete 3-D transform with the discrete analytical geometry theory by the construction of 3-D discrete analytical lines in the Fourier domain. We propose two types of 3-D discrete lines: 3-D discrete radial lines going through the origin defined from their orthogonal projections and 3-D planes covered with 2-D discrete line segments. These discrete analytical lines have a parameter called arithmetical thickness, allowing us to define a 3-D DART adapted to a specific application. Indeed, the 3-D DART representation is not orthogonal, It is associated with a flexible redundancy factor. The 3-D DART has a very simple forward/inverse algorithm that provides an exact reconstruction without any iterative method. In order to illustrate the potentiality of this new discrete transform, we apply the 3-D DART and its extension to the Local-DART (with smooth windowing) to the denoising of 3-D image and color video. These experimental results show that the simple thresholding of the 3-D DART coefficients is efficient.

This paper introduces a new strategy to describe wavelet sub-bands in the scope of texture characterization. While most wavelet-based texture features found in the literature do not use spatial information contained in the wavelet domain, the new approach suggests to represent sub-bands of a texture by shape parameters. This new technique is called wavelet geometrical features (WGF), and is coming from a multiresolution extension of Chen's statistical geometrical features (SGF). Experiments on the full Brodatz database show the efficiency of the WGF over the SGF and the traditional wavelet energy signature.

The paper presents a watermarking technique, specific to color images. The insertion and the detection are based on the 2D discrete wavelet transform, applied on each color component. Three color vectors are extracted from the wavelet coefficients. The insertion consists of modifying one vector for each location, with regard to the bit value and the vector triplet. The mark is extracted without the original image, by observing the scheme of each vector triplet. This new method has been shown to be resistant to JPEG compression, median filtering and noise adding. Moreover, each insertion is weighted by an adaptive strength, obtained by a retroactive process between the original and the watermarked images. This process allows optimizing the compromise between invisibility and robustness, considering the local image color content.

In this paper, we present a fast implementation of the 3D ridgelet transform based on discrete analytical 3D lines: the 3D discrete analytical ridgelet transform (DART). This transform uses the Fourier strategy (the projection-slice formula) for the computation of the associated discrete Radon transform. The innovative step of the DART is the construction of 3D discrete analytical lines in the Fourier domain, that allows a fast perfect backprojection. These discrete analytical lines have a parameter called arithmetical thickness, allowing us to define a DART adapted to a specific application. A denoising application is presented.

We propose a simple and new way of defining the directional wavelet packets basis. This new decomposition is based on a discrete rotation. To avoid information loss or modification (in order to keep the orthogonality property) we use a new one-to-one discrete rotation. This rotation is based on three discrete shear transformations. This new oriented basis does not divide the 2D Fourier plane into square regions. Instead of horizontal and vertical favourite directions, our directional scheme treats different arbitrary directions.

We study the problem of estimating the log-spectrum of a stationary Gaussian time series by thresholding the wavelet coefficients. We propose the use of the undecimated wavelet transform to denoise the log-periodogram. For this, we review a denoising method based on undecimated wavelet transform, and we propose a level-dependent threshold which considers that one undecimated scale has N/b coefficients "repeating" b times. The result corresponds to the average of all log-peridogram circulant shifts denoised by a decimated wavelet transform. The purpose of this undecimated thresholding is to make the reconstructed log-spectrum as nearly noise-free as possible, but with a keep of all small frequential components. Since the wavelet denoising method can be generalized to images, we develop an estimation technique of the 2D log-spectrum based on 2D undecimated wavelet. We derive a new technique, easy to apply, which gives information about the 2D frequential components of an image.

In this article, we propose a simple algorithm allowing the research of stationary partitions in locally stationary 2D processes, by use of adapted local bases like the local 2D DCT. The algorithm aims at finding some windows for which the width is adapted to the rate of change in the 2D spectrum. For this, we define an anisotropic Malvar's decomposition with a weak complexity and we propose a robust estimation of the difference in spectra, based on a denoising of the 2D log-periodogram with the help of the undecimated wavelet transform. Then, we consider the "symmetrical case" in the best partition selection method and a post-treatment that permits us to deal with the 2D nonuniform partitions is introduced. Our algorithm obtains better results than the classical Malvar's decomposition. It must be considered as a first treatment, which selects a global segmentation.