Pierre Baylou

Adaptive nonlinear filters for 2D and 3D image enhancement

Abstract Unsharp masking method is a popular approach for image enhancement, in which a highpass version of an image is added to the original one. This method is easy to run, but is very sensitive to noise. Suppressing noise is generally performed with lowpass filters, and leads to edge blurring. So, an approach which is a combination of a nonlinear lowpass and highpass filters is proposed. These filters are based on an adaptive filter mask. We demonstrate that this approach performs noise reduction as well as edge enhancement. It also improves the contrast enhancement in comparison with other methods. These results are illustrated by processing blurred and noisy images. The method is then extended for 3D data processing and used on 3D seismic images.

Estimating local multiple orientations

This paper focuses on the estimation of local orientation in an image where several orientations exist at the same location and at the same scale. Within this framework, Isotropic and Recursive Oriented Network (IRON), an operator based on an oriented network of parallel lines is introduced. IRON uses only a few parameters. Beyond the choice of a specific line homogeneity feature, the size and the shape of the network can be tuned. These parameters allow us to adapt our operator to the image studied. The implementation we propose for the network is recursive, relying on the rotation of the image instead of the rotation of the operator. IRON can proceed on a small computing support, and thus provides a local estimation of orientations. Herein, we test IRON on both synthetic and real images. Compared to some other orientation estimation methods such as Gabor filters or Steerable filters, our operator detects multiple orientations with both better accuracy and noise robustness, at a competitive computational cost thanks to its recursivity. Moreover, IRON offers better selectivity, particularly at small scale.

Active contours network to straighten distorted text lines

We introduce a new method to straighten text lines. The general approach consists of using an active contour network based on an analytical model. First, we propose a method based on Bezier curves; then we present a model that requires cubic B-splines which leads to more accurate results. Finally, we propose to automate the initialization by using an approach based on a particle system. The proposed methods are illustrated on real images of curved text.

Image segmentation by a dynamic thresholding using edge detection based on cascaded uniform filters

Abstract A new method for gray level thresholding is presented in this paper. First, the edge pixels can be localized by the difference between two cascaded uniform filters with different scales. Then, considering the gray level value of an edge pixel as the ideal threshold for its neighborhood, a dynamic threshold function will be constructed by a dilation process from these primitive thresholds. The experimental results we provide in this paper show that this technique is very effective in segmenting images taken under non-uniform illuminations.

A new adaptive framework for unbiased orientation estimation in textured images

This paper focuses on directional texture analysis. We propose a new approach for orientation estimation. This approach hinges on two classes of convolution masks, i.e. the gradient and the valleyness operators. We provide a framework for their optimization regarding bias reduction and noise robustness. As the gradient and the valleyness operators are complementary, we propose a combination named GV-JOE. This combination consists in using the gradient on inflexion pixels, the valleyness on crests and valleys, and a linear mixture of both elsewhere. We implement an adaptive selection of the size of our operators, in order to take into account the variations of the texture scale in the image. We apply our approach both on synthetic and natural textures. These experiments show that, when used separately, both classes of operators are more accurate than classical derivative approaches. In noisy cases, the GV-JOE implementation improves the robustness of our operators without affecting their accuracy. Moreover, compared to well-known orientation estimators, it gives the best estimates in the most difficult cases i.e. for high-frequency textures and low SNR.

Mixed anisotropic diffusion

This paper deals with the filtering and the enhancement of strongly oriented patterns. We propose a new filter combining scalar and tensor based diffusivities. The computation of an orientation confidence allows its to choose the best strategy to locally diffuse the gray levels. We show that this approach overcomes some drawbacks of the classical methods like corner smoothing or pinhole effect. The proposed method was applied on digital images of engravings. This type of images contain strongly oriented patterns and a large amount of details that have to be preserved during the diffusion.

Curvature of oriented patterns: 2-D and 3-D estimation from differential geometry

This paper presents a method that estimates curvature of both 2-D and 3-D oriented patterns at different scales even if the structures are very close. Assuming that a structure is locally defined by an implicit isointensity contour, the curvature is obtained by a direct computation stemming from the differential geometry. The method is applied to curvature estimation of seismic image for geological interpretation.

A new 2-D spectrum estimate using multichannel AR approach of 2-D fast RLS algorithms

In the framework of high resolution 2-D spectrum analysis, a new multichannel approach called harmonic mean horizontal vertical (HMHV) is proposed. It is based on 2-D fast recursive least squares (2-D FRLS) algorithms and their use for the computation of causal 2-D autoregressive (AR) parameters. This HMHV spectrum presents the following three main advantages on the 2-D spectrum estimated by the harmonic mean (HM) of the 2-D AR first and second quarter plane supports (QP1 and QP2) spectrum estimates: first, it presents the same biases and variances of estimation for the horizontal and vertical frequency components and improves in many cases the variances obtained with the HM method. Secondly, the single peak area (SPA) of the HMHV estimate is quite circular although the HM one looks like a skewed square indicating the existence of a best direction for the separation of two sinusoids. Thirdly, the new estimate presents less spurious peaks. This paper sums up the calculation of the different spectrum estimates and the experiments which lead to the conclusions.

Robust nonlinear contrast enhancement filters

This paper is devoted to a new adaptive class of nonlinear contrast enhancement filters. These filters are referred to as nonstationary filters because the filter mask depends on the local pixel values. First we show how to find the adaptive filter mask, i.e. the set of surrounding pixels the more similar, in a sense which is defined, to the current pixel. Then we define a new class of quadratic filters based on the so defined filter mask. Finally, processing of blurred and noisy images by these algorithms demonstrates the contrast enhancement improvement in comparison to other quadratic filters.

Extension of the minimal path searching for structure recovery

A generalization of the notion of minimal path is presented in order to retrieve the structure of tree-like objects when using grayscale images. More particularly, we extend the notion of the regularization term used in an active contour model. As the energy function proposed does not strictly increase with the distance, the path propagates inside the object, even in the case of a curved object. Then, a classical classification algorithm is proposed to generate the cartography of a tree shaped-object structure. This algorithm is computed using the dissimilarity matrix obtained by searching the minimal paths between a set of given start points. Finally, we show the results obtained on aerial and biomedical images.