Spectral Reflectance Reconstruction Using Fuzzy Logic System Training: Color Science Application.

In this work, we address the problem of spectral reflectance recovery from both CIEXYZ and RGB values by means of a machine learning approach within the fuzzy logic framework, which constitutes the first application of fuzzy logic in these tasks. We train a fuzzy logic inference system using the Macbeth ColorChecker DC and we test its performance with a 130 sample target set made out of Artist's paints. As a result, we obtain a fuzzy logic inference system (FIS) that performs quite accurately. We have studied different parameter settings within the training to achieve a meaningful overfitting-free system. We compare the system performance against previous successful methods and we observe that both spectrally and colorimetrically our approach substantially outperforms these classical methods. In addition, from the FIS trained we extract the fuzzy rules that the system has learned, which provide insightful information about how the RGB/XYZ inputs are related to the outputs. That is to say that, once the system is trained, we extract the codified knowledge used to relate inputs and outputs. Thus, we are able to assign a physical and/or conceptual meaning to its performance that allows not only to understand the procedure applied by the system but also to acquire insight that in turn might lead to further improvements. In particular, we find that both trained systems use four reference spectral curves, with some similarities, that are combined in a non-linear way to predict spectral curves for other inputs. Notice that the possibility of being able to understand the method applied in the trained system is an interesting difference with respect to other 'black box' machine learning approaches such as the currently fashionable convolutional neural networks in which the downside is the impossibility to understand their ways of procedure. Another contribution of this work is to serve as an example of how, through the construction of a FIS, some knowledge relating inputs and outputs in ground truth datasets can be extracted so that an analogous strategy could be followed for other problems in color and spectral science.

Graphs based methods for simultaneous smoothing and sharpening.

We present two new methods for simultaneous smoothing and sharpening of color images: the GMS3 (Graph Method for Simultaneous Smoothing and Sharpening) and the NGMS3(Normalized Graph-Method for Simultaneous Smoothing and Sharpening). They are based on analyzing the structure of local graphs computed at every pixel using their respective neighbors. On the one hand, we define a kernel-based filter for smoothing each pixel with the pixels associated to nodes in its same connected component. On the other hand, we modify each pixel by increasing their differences with respect to the pixels in the other connected components of those local graphs. Our approach is shown to be competitive with respect to other state-of-the-art methods that simultaneously manage both processes.•We provide two methods that carry out the process of smoothing and sharpening simultaneously.•The methods are based on the analysis of the structure of a local graph defined from the differences in the RGB space among the pixels in a 3 × 3 window.•The parameters of the method are adjusted using both observers opinion and the well-known reference image quality assessment BRISQUE (Blind/Referenceless images spatial quality Evaluator) score.

Method to determine the degrees of consistency in experimental datasets of perceptual color differences.

We propose a fuzzy method to analyze datasets of perceptual color differences with two main objectives: to detect inconsistencies between couples of color pairs and to assign a degree of consistency to each color pair in a dataset. This method can be thought as the outcome of a previous one developed for a similar purpose [J. Mod. Opt.56, 1447 (2009)JMOPEW0950-034010.1080/09500340902944038], whose performance is compared with the proposed one. In this work, we present the results achieved using the dataset employed to develop the current CIE/ISO color-difference formula, CIEDE2000, but the method could be applied to any dataset. Specifically, in the mentioned dataset, we find that some couples of color pairs have contradictory information, which can interfere in the successful development of future color-difference formulas as well as in checking the performance of current ones.

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.

Robustifying vector median filter.

This paper describes two methods for impulse noise reduction in colour images that outperform the vector median filter from the noise reduction capability point of view. Both methods work by determining first the vector median in a given filtering window. Then, the use of complimentary information from componentwise analysis allows to build robust outputs from more reliable components. The correlation among the colour channels is taken into account in the processing and, as a result, a more robust filter able to process colour images without introducing colour artifacts is obtained. Experimental results show that the images filtered with the proposed method contain less noisy pixels than those obtained through the vector median filter. Objective measures demonstrate the goodness of the achieved improvement.

Fuzzy peer groups for reducing mixed gaussian-impulse noise from color images.

The peer group of an image pixel is a pixel similarity-based concept which has been successfully used to devise image denoising methods. However, since it is difficult to define the pixel similarity in a crisp way, we propose to represent this similarity in fuzzy terms. In this paper, we introduce the fuzzy peer group concept, which extends the peer group concept in the fuzzy setting. A fuzzy peer group will be defined as a fuzzy set that takes a peer group as support set and where the membership degree of each peer group member will be given by its fuzzy similarity with respect to the pixel under processing. The fuzzy peer group of each image pixel will be determined by means of a novel fuzzy logic-based procedure. We use the fuzzy peer group concept to design a two-step color image filter cascading a fuzzy rule-based switching impulse noise filter by a fuzzy average filtering over the fuzzy peer group. Both steps use the same fuzzy peer group, which leads to computational savings. The proposed filter is able to efficiently suppress both Gaussian noise and impulse noise, as well as mixed Gaussian-impulse noise. Experimental results are provided to show that the proposed filter achieves a promising performance.

A new fuzzy color correlated impulse noise reduction method.

A new impulse noise reduction method for color images is presented. Color images that are corrupted with impulse noise are generally filtered by applying a grayscale algorithm on each color component separately or using a vector-based approach where each pixel is considered as a single vector. The first approach causes artefacts especially on edge and texture pixels. Vector-based methods were successfully introduced to overcome this problem. Nevertheless, they tend to cluster the noise and to receive a lower noise reduction performance. In this paper, we discuss an alternative technique which gives a good noise reduction performance while much less artefacts are introduced. The main difference between the proposed method and other classical noise reduction methods is that the color information is taken into account to develop (1) a better impulse noise detection method and (2) a noise reduction method that filters only the corrupted pixels while preserving the color and the edge sharpness. Experimental results show that the proposed method provides a significant improvement on other existing filters.