Segmentation of lung parenchyma from the chest computed tomography is an important task in analysis of chest computed tomography for diagnosis of lung disorders. It is a challenging task especially in the presence of peripherally placed pathology bearing regions. In this work, we propose a segmentation approach to segment lung parenchyma from chest. The first step is to segment the lungs using iterative thresholding followed by morphological operations. If the two lungs are not separated, the lung junction and its neighborhood are identified and local thresholding is applied. The second step is to extract shape features of the two lungs. The third step is to use a multilayer feed forward neural network to determine if the segmented lung parenchyma is complete, based on the extracted features. The final step is to reconstruct the two lungs in case of incomplete segmentation, by exploiting the fact that in majority of the cases, at least one of the two lungs would have been segmented correctly by the first step. Hence, the complete lung is determined based on the shape and region properties and the incomplete lung is reconstructed by applying graphical methods, namely, reflection and translation. The proposed approach has been tested in a computeraided diagnosis system for diagnosis of lung disorders, namely, bronchiectasis, tuberculosis, and pneumonia. An accuracy of 97.37 % has been achieved by the proposed approach whereas the conventional thresholding approach was unable to detect peripheral pathology-bearing regions. The results obtained prove to be better than that achieved using conventional thresholding and morphological operations.

Summary Proposed is a discriminative feature modeling technique in three orthogonal planes (TOP) for human action recognition (HAR). Pyramidal histogram of orientation gradient‐TOP (PHOG‐TOP) and dense optical flow‐TOP (DOF‐TOP) techniques are utilized for salient motion estimation and description to represent the human action in a compact but distinct manner. The contribution of the work is to explicitly learn the gradual change of visual patterns using fusion of PHOG‐TOP and DOF‐TOP technique to discover the nature of the action. With this feature representation, dimensionality reduction is achieved by deep stacked autoencoders. The encoded feature representation is used in long short term memory (LSTM) classification for HAR. Experiments with better recognition rate demonstrate the discriminative power of the proposed descriptor. Moreover, the proposed modeling and LSTM classification outperforms the state of art methods for HAR.