Neural network texture segmentation of satellite images of woodlands using the U-net model

With the advent of space equipment that allows obtaining panchromatic images of ultra-high spatial resolution (< 1 m) there was a tendency to develop methods of thematic processing of aerospace images in the direction of joint use of textural and spectral features of the objects under study. In this paper, we consider the problem of classification of forest canopy structures based on textural analysis of multispectral and panchromatic images of Worldview-2. Traditionally, a statistical approach is used to solve this problem, based on the construction of distributions of the common occurrence of gray gradations and the calculation of statistical moments that have significant regression relationships with the structural parameters of stands. An alternative approach to solving the problem of extracting texture features is based on frequency analysis of images. To date, one of the most promising methods of this kind is based on wavelet scattering. In comparison with the traditionally applied approaches based on the Fourier transform, in addition to the characteristic signal frequencies, the wavelet analysis allows us to identify characteristic spatial scales, which is fundamentally important for the textural analysis of spatially inhomogeneous images. This paper uses a more general approach to solving the problem of texture segmentation using the convolutional neural network U-net. This architecture is a sequence of convolution-pooling layers. At the first stage, the sampling of the original image is lowered and the content is captured. At the second stage, the exact localization of the recognized classes is carried out, while the discretization is increased to the original one. The RMSProp optimizer was used to train the network. At the preprocessing stage, the contrast of fragments is increased using the global contrast normalization algorithm. Numerical experiments using expert information have shown that the proposed method allows segmenting the structural classes of the forest canopy with high accuracy.

A Hybrid Multiresolution Texture Segmentation and its Application to Ikonos Satellite Imagery

In this thesis a method for segmenting textured images using Gabor filters is presented. One of the most recent approaches for texture segmentation and analysis is multi-channel filtering. There are several applicable choices as filter banks which are used for textured images. Gaussian filters modulated by exponential or by sinusoidal filters, known as Gabor filters, have been proven to be very usefyl for texture analysis for the images containing specific frequency and orientation characteristics. Resembling the human visual cortical cells, Gabor function is a popular sub-band filter for multi-channel decompositon. Optimum joint spatial/spatial frequency uncertainty principle and its ability to recognize and pass specific frequencies and orientations are attributes of Gabor filter that make it more attractive. Gabor function with these attributes could simulate the task of simple visual cells in the cortex. Gabor function has several parameters that determine the sub-band Gabor filter and must be determined accurately to extract the features precisely for texture discrimination. A wide selection range for each parameter exists and many combinations of these parameters are possible. Accurate selection and combination of values for the parameters are of crucial importance. Hence a difficult goal is minimizing the number of filters. On the other hand a variety of approaches of texture analysis and recognition have been presented in remote sensing applications, including land cover/land use classification and urban scene segmentation. With the avaiability of very high-resolution commercial satellite imagery such as IKONOS, it is possible to obtain detailed information on urban land use and change detection that are of particular interest to urban and regional planners. In this thesis considering the attributes of human visual system, a hybrid algorithm is implemented using multi-channel decomposition by Gabor filter bank for feature extraction in conjunction with Artificial Neural Networks for both feature reduction and texture segmentation. Three approaches are implemented to optimize Gabor filter bank for image segmentation. Eventually the proposed method is successfully applied for segmentation of IKONOS satellite images.

A Hybrid Multiresolution Texture Segmentation and its Application to Ikonos Satellite Imagery

In this thesis a method for segmenting textured images using Gabor filters is presented. One of the most recent approaches for texture segmentation and analysis is multi-channel filtering. There are several applicable choices as filter banks which are used for textured images. Gaussian filters modulated by exponential or by sinusoidal filters, known as Gabor filters, have been proven to be very usefyl for texture analysis for the images containing specific frequency and orientation characteristics. Resembling the human visual cortical cells, Gabor function is a popular sub-band filter for multi-channel decompositon. Optimum joint spatial/spatial frequency uncertainty principle and its ability to recognize and pass specific frequencies and orientations are attributes of Gabor filter that make it more attractive. Gabor function with these attributes could simulate the task of simple visual cells in the cortex. Gabor function has several parameters that determine the sub-band Gabor filter and must be determined accurately to extract the features precisely for texture discrimination. A wide selection range for each parameter exists and many combinations of these parameters are possible. Accurate selection and combination of values for the parameters are of crucial importance. Hence a difficult goal is minimizing the number of filters. On the other hand a variety of approaches of texture analysis and recognition have been presented in remote sensing applications, including land cover/land use classification and urban scene segmentation. With the avaiability of very high-resolution commercial satellite imagery such as IKONOS, it is possible to obtain detailed information on urban land use and change detection that are of particular interest to urban and regional planners. In this thesis considering the attributes of human visual system, a hybrid algorithm is implemented using multi-channel decomposition by Gabor filter bank for feature extraction in conjunction with Artificial Neural Networks for both feature reduction and texture segmentation. Three approaches are implemented to optimize Gabor filter bank for image segmentation. Eventually the proposed method is successfully applied for segmentation of IKONOS satellite images.

An image-computable model of attention and texture segmentation

ABSTRACTAttention can facilitate or impair texture segmentation, altering whether objects are isolated from their surroundings in visual scenes. We simultaneously explain several empirical phenomena of texture segmentation and its attentional modulation with a single image-computable model. At the model’s core, segmentation relies on the interaction between sensory processing and attention, with different operating regimes for involuntary and voluntary attention systems. Model comparisons were used to identify computations critical for texture segmentation and attentional modulation. The model reproduced (i) the central performance drop, which is the parafoveal advantage for segmentation over the fovea, (ii) the peripheral improvements and central impairments induced by involuntary attention and (iii) the uniform improvements across eccentricity by voluntary attention. The proposed model reveals distinct functional roles for involuntary and voluntary attention and provides a generalizable quantitative framework for predicting the perceptual impact of attention across the visual field.