4. Methods of Filtering and Texture Segmentation of Multicomponent Images

Some modern video systems, for example, remote sensing systems analyze the multicomponent images. Limitations of on-board technical and energy resources and video data transmission by low power and over long distance lead to strong image distortions. The filtering is used to recover the distorted by noise images for subsequent tasks of image processing, such as detection of texture regions and objects of interest, estimations of their parameters, classification, and recognition. Multicomponent images can be represented as the multidimensional signals and have significantly greater statistical redundancy than one-component images. This redundancy would be appropriate to improve a quality of image restoration. Special cases of multicomponent images are color RGB images, each color component of which is a g-bit digital halftone image. The nature of the statistical relationship between elements within the digital halftone image and among the elements of color components allows to suggest an approximation for 3D color images using a Markov chain with several states and for bit binary image applying a 3D Markov chain with two states. The proposed filtering method is based on an approximation the multicomponent images using a 3D Markov chain and on an efficient use of statistical redundancy of multicomponent images. This method requires small computational resources and is effective with signal-to-noise ratio at the input of receiver up to –9 dB. Real images have areas with varying degrees of detail and different statistical characteristics. The authors propose to improve the accuracy of the statistical characteristics of each local region within an image and between the color components to improve a quality of the reconstructed image. A sliding window is used to estimate the local statistical characteristics of an image. The proposed method allows to detect the small objects and contours of objects more accurately in image distorted by white Gaussian noise. A method of texture regions’ detection on the reconstructed images based on Markov random fields is proposed. An estimate of the probability of a transition between image elements is used as the texture feature. The method efficiently detects the texture regions with different statistical characteristics and makes it possible to reduce the computational costs.