The Cherenkov Telescope Array is the future of ground-based gamma-ray astronomy. Its first prototype telescope built on-site, the Large Size Telescope 1, is currently under commissioning and taking its first scientific data. In this paper, we present for the first time the development of a full-event reconstruction based on deep convolutional neural networks and its application to real data. We show that it outperforms the standard analysis, both on simulated and on real data, thus validating the deep approach for the CTA data analysis. This work also illustrates the difficulty of moving from simulated data to actual data.

Smartphones bring a new way to scan and digitalize written documents by taking pictures. This enables new document analysis applications to emerge. As a counterpart, unsupervised document capturing brings new challenges mainly related to target document localization and high quality text recognition. In this context, this work addresses automatic sale receipt understanding in an industrial context. It relies on the extraction of accurate and essential consumption data even with low quality receipt captures. We propose a tool chain that combines Deep Neural Networks and traditional image processing to ensure accurate automatic data extraction. The proposed workflow is evaluated globally by the analysis of the quality of the text recognition at the end of the processing.

Recently, a variety of approaches have been enriching the field of remote sensing (RS) image processing and analysis. Unfortunately, existing methods remain limited to the rich spatiospectral content of today's large data sets. It would seem intriguing to resort to deep learning (DL)-based approaches at this stage with regard to their ability to offer accurate semantic interpretation of the data. However, the specificity introduced by the coexistence of spectral and spatial content in the RS data sets widens the scope of the challenges presented to adapt DL methods to these contexts. Therefore, the aim of this paper is first to explore the performance of DL architectures for the RS hyperspectral data set classification and second to introduce a new 3-D DL approach that enables a joint spectral and spatial information process. A set of 3-D schemes is proposed and evaluated. Experimental results based on well-known hyperspectral data sets demonstrate that the proposed method is able to achieve a better classification rate than state-of-the-art methods with lower computational costs.

With the rapid development of Remote Sensing acquisition techniques, there is a need to scale and improve processing tools to cope with the observed increase of both data volume and richness. Among popular techniques in remote sensing, Deep Learning gains increasing interest but depends on the quality of the training data. Therefore, this paper presents recent Deep Learning approaches for fine or coarse land cover semantic segmentation estimation. Various 2D architectures are tested and a new 3D model is introduced in order to jointly process the spatial and spectral dimensions of the data. Such a set of networks enables the comparison of the different spectral fusion schemes. Besides, we also assess the use of a “noisy ground truth” (i.e. outdated and low spatial resolution labels) for training and testing the networks.

In this paper, a new model for color image enhancement based on diffusion-shock filter coupling is presented. The proposed method is based on the consideration of single vectors of the gradient magnitude and the second derivatives as a manner to relate the processing of the different color components of the image. This solution avoids the apparition of color artefacts in the enhanced image.

We propose a straightforward intensity-based dissolve detection method which is able to cope with the particular constraints of the artistic animated movie domain. It uses the hypothesis that during a dissolve, the amount of fading-out and fading-in pixels should be high. Instead of just applying a global threshold, as most of the existing approaches do, we use a twin-threshold approach coped with shape analysis of the signal. This approach allows to reduce false detections caused by steep intensity fluctuations as well as to retrieve dissolves caught up in other visual effects. Experimental tests show that whether classic approaches tends to fail, the pro posed method provides a good detection even for very complex content setups (e.g. global movement, temporal discontinuity).

We address a particular case of video genre classification, namely the classification of animated movies. This task is achieved using two categories of content descriptors, temporal and color based, which are adapted to this particular content. Temporal descriptors, like rhythm or action, are quantifying the perception of the action content at different levels. Color descriptors are determined using color perception which is quantified in terms of statistics of color distribution, elementary hues, color properties (e.g. amount of light colors, cold colors, etc.) and color relationship. The potential of the proposed descriptors to the classification task has been proved through experimental tests conducted on more than 159 hours of video footage. Despite the high diversity of the video material, the proposed descriptors achieve an average precision and recall ratios up to 90% and 92%, respectively, and a global correct detection ratio up to 92%.

Multimedia documents are increasingly numerous. Their efficient management requires tools to provide services that measure up to users' expectations, based on the contents of these voluminous document databases. This implies a number of challenges. Although we can extract highly symbolic concepts from texts, a wide semantic gap appears when processing images and sound. Thus, we propose using data fusion to capture the semantic field of the elements extracted from these media. On the other hand, if creating ontology makes it possible to manage the concept of "trade" in texts, additional data can be exploited from document contents. In the first instance, we suggest bringing together concepts and words through data fusion.

In this paper an improved cut detection algorithm, adapted to the segmentation of animation movies, is proposed. As color is a major feature of animation movies (each movie has its own particular color distribution) the proposed algorithm applies second order derivatives on Euclidean distances between color histograms of frames quadrants in order to improve the cut detection. For the frame classification, an automatic threshold estimation is proposed. Also, in order to reduce false detections, we propose an algorithm to detect an effect specific to animation movies, named "short color change" (i.e. thunders, lightening). The resulting method achieved better results compared to the classical histogram-based and motion-discontinuity based approaches, as shown by tests conducted on several animation movies

This paper presents a method for the automatic generation of animation movie trailers. First, the movie is divided into shots, by detecting the video transitions (cuts, fades and dissolves) and an animation movie specific color effect named short color change or SCC (i.e. explosions, thunders). The movie action content is further highlighted by analyzing the movie at two different granularity levels. First, an inter-shot analysis is performed by measuring the video transition temporal distribution. As repetitive shot changes are related to action, we define as an action shot, a movie segment containing a high shot change ratio. On the other hand, an inter-frame analysis is performed: for each shot within an action shot, an histogram of cumulative inter-frame distances is computed to serve as a measure of the frame spatial activity. Repetitive color changes are also related to action. Since movie trailers only show some of the most attractive movie scenes, the proposed trailer is a moving-image abstract computed on the retained action shots. It provides the user with a compact and efficient representation of the movie action content. The proposed approach was tested on several animation movies.

In this paper we focus on the problem of image segmentation by color classification. We present a robust agglomerating clustering algorithm based on a cluster validity criteria derived from fuzzy partitions. The result is a simplified segmentation having a small number of large regions. The interest of the proposed method is that it requires a single parameter and that the computational complexity is very low.

We focus on the problem of partial queries on content-based image retrieval. We first address the problem of region of interest delimitation through image segmentation by color clustering. We present a robust agglomerating clustering algorithm based on cluster validity criteria derived from fuzzy partitions. We secondly address the problem of color similarity measurement between partial queries using the clustering information obtained from the segmentation stage. We present results to assess the effectiveness of both the segmentation algorithm and retrieval process on a large collection of real-world images.

A segmentation can be represented by a graph where each vertex corresponds to a region and where the arc between two vertices (i.e. two regions) denotes the connectivity between the two regions. Each vertex and each arc being characterized by a set of attributes (color attributes, geometrical attributes, ...), the first aim of the paper is to define a symbolic description for each attribute. Then, by using rule basis, the similarity between two adjacent regions is also described in a symbolic way. Finally, according to this similarity definition the graph is reduced to get a simplified segmentation. In the presented applications, it is shown that the simplification can be performed in different ways, depending on a the aim of the analysis.

The problem of detection of road networks in Synthetic Aperture Radar (SAR) images is addressed. This task is important for the registration of SAR images to existing maps and for perfect superposition of multitemporal SAR images. SAR image analysis is a difficult task because of speckle noise and textures and low signal to noise ratios. The paper focuses on the comparison of filter performances in terms of road extraction. The study points out the interest of nonlinear adaptive filtering techniques. On the other hand, it is shown that techniques dedicated to speckle noise filtering are not always the most suited to road network extraction.

In this paper, the hardware implementation of a vector median like filter is proposed. Firstly, the extension of median filtering to the case of multicomponent images is addressed. It requires the definition of an appropriate vector ordering scheme, which is not straightforward. The proposed solution is based on the bit mixing approach, which consists in a revertible scalar coding of the vector input data, followed by a simple scalar processing of the encoded data. The filtered multicomponent image is obtained after decoding. An application to color images is presented. The second part of this work is the real time implementation of this filter. An FPGA design is proposed and the detailed structure of this implementation is presented.

This paper is concerned with the automatic detection of linear features in SAR satellite data, with application to road network extraction. After a directional prefiltering step, a morphological line detector is presented. To improve the detection performances, the results obtained on multitemporal data are fused. Different fusion strategies involving different fusion operators are then presented. Since extensions of classical set union and intersection do not lead to satisfactory results (the corresponding operators are either too indulgent or too severe), the first strategy consists of fusing the data using a compromise operator. The second strategy consists of fusing the results computed with two operators that have opposite properties, in order to obtain a final intermediate result. Thanks to the wide range of properties they provide, fuzzy operators are used to test and compare these two fusion strategies on real ERS-1 multitemporal data.

A method for color image segmentation processing is presented. This method is based upon several linguistic rules which built a symbolic cooperation between symbolic representation of hue, chroma and intensity features. This fuzzy segmentation is compared with a technique using a fuzzy C-means algorithm, and the influence of the choice of the rules is analyzed.

In the hue saturation intensity (HSI) space, a color edge detection approach based on fuzzy if-then rules is presented. This method avoids the difficulty to combine hue and intensity features. Noise sensitivity and edges localization are studied on synthetic images. Application on real images is given. The results show a good edge localization and a noise sensitivity.

In the hue saturation intensity (HSI) space, a hue difference taking into account the hue relevance is defined. A hue gradient operator is build up with this hue difference. Two color edge detectors are then proposed. These detectors mix in a different way H,S,I gradients. Experimental results are presented.>