An Infrared Image Preprocessing Method Using Improved Adaptive Neighborhoods
This work proposes the use of an adaptive neighborhood procedure to extract local statistical properties of images in order to improve a speckle noise “Maximum a Posteriori ”(MAP) filtering performance. The strategy consists in growing the local statistically homogeneous area near the pixel in order to estimate its MAP filtering parameters. Measures evaluating both the signal-to-noise improvement and resolution loss due to filtering are computed. The use of region growing is investigated as a promising approach compared to a fixed size and shape neighborhood. 1-Introduction The analysis of synthetic aperture radar (SAR) images requires in a first step of processing to reduce the speckle noise in order to improve the overall visual aspect of the image. The use of SAR images filtering improves the segmentation and classification considerably because filtered images are easier to classify. This kind of noise degrades images generated by coherent signal sources, such as radar signals. In speckle noise filtering there is a compromise between the smoothing of the homogeneous areas and the edges and detail preservation. The basic requirements of those filtering algorithms are: to provide a large amount of speckle noise reduction in homogeneous areas and to prevent edges and detail (resolution) blurring. Various algorithms have been proposed to restore noisy images using local statistics, including those proposed by Lee  , Kuan  and Frost  which use fixed size and shape local neighborhoods to adapt the filtering techniques on a small square area around the noisy pixel. The adaptive strategy is to select a neighborhood of the pixel which is suitable for calculating statistical measures (mean, variance, for example) to update the central pixel according to a filter based on those statistics. Medeiros et al.  developed a speckle noise filtering technique that combines the “Maximum a Posteriori” estimation and the k-means clustering algorithm. The k-means over Changle Li’s coefficient  is used to classify the noisy image in regions of homogeneous statistics which is used as a guide for choosing the best window size for parameter estimation . The dependence of results on the shape and dimensions of the processing window is a well-known problem in image processing algorithms. According to the main characteristics of the image under analysis or the processing objective, the window can be carefully designed in linear or nonlinear algorithms in order to reduce the bias error . However, it is well known that large filtering window sizes cause resolution degradation. So, the requirement for an edge preserving filter is to reduce the noise variance while preserving edges and details. One of the contibutions of this paper is to propose the use of adaptive neighborhoods within the MAP approach, investigating the benefits and drawbacks. For the proposed algorithm, we aim to find a better subset around each pixel (without fixed shape and size) whose respective statistical information is more suitable to perform the MAP filtering. These subsets of pixels are expected to form statistically homogeneous regions on which the local statistics (mean and variance) used in the adaptive MAP filters are estimated. The estimation of the parameters is based only on those pixels of the grown region that are supposed to have the same statistics as the central pixel to be filtered. This work investigates the improvements that can be achieved by adopting such a methodology. The MAP filtering based on region growing approach is adapted to favor the speckle noise filtering in homogeneous areas. This paper is organized as follows: section 2 presents a brief review of the multiplicative speckle model and describes the Kuan’s filter  used to be compared with the proposed algorithm. In the third section the MAP technique combined with the region growing approach is described. In section 4 there is a description of the quality measures used to quantify the reduction of the speckle in the MAP filtering algorithm. Section 5 includes the simulations results, discussions about the advantages and limitations of this approach and concluding remarks.