Showing papers by "Pritee Khanna published in 2013"

ROI segmentation using Local Binary Image

Abstract: Segmentation of ROI is an important and challenging task in the development of CAD system for the detection of breast cancer. This work proposes a Local Binary Image (LBI) to segment the ROI from the mammogram patches. The key idea is to use textural properties of mammogram patches for representing salient micro-patterns of the masses and preserving the spatial information at the same time. Corresponding to the patch, LBI is the binary image where the value 1 represents the presence of texture in the patch. Using LBI the threshold value is identified which is used to extract the mask image. Once the mask image is generated boundary is plotted to trace suspicious area in the patch. The efficiency of the proposed method is tested on a dataset of 819 suspicious patches from the IRMA reference database. The experimental results achieved that the proposed LBI method has successfully attained the value 0.934 for Quality measure.

Extended Axiomatic Design and Computational Support to Design for Aesthetics

Abstract: The research describes the extension of the Axiomatic Design model to incorporate the aesthetic design as the customer requirement. It also proposes a computational model to support the formalization of aesthetic design in industrial products. The methodology takes into account the cognition process during the design generation and captures this behavior in a group theoretic structure. This approach leads to application of Axiomatic Design paradigm to the domain of the aesthetics. The proposed framework is implemented and validated by taking a design case of the consumer products.

Face verification and identification using DCT-NNDA and SIFT with score-level fusion

Abstract: This paper proposes a new face verification and identification system based on the fusion of global and local features of face. DCT is used to extract global features from face images. A non-parametric discrimination method, NNDA is applied on the global features to make them more compact within the class clusters; while separating among the class clusters. The effects of expression variations are removed by DWT. It is found that DCT-NNDA is robust to small noisy (blurred) faces, but its performance degrades gradually for variations in scale, rotation, expression and pose. These issues are resolved using local feature extraction through SIFT. The work utilises strengths of DCT-NNDA and SIFT along with score level fusion. The proposed method is robust to scale, small noise, pose, expression, and illumination variations. The proposed system achieves 1% and 0.2% EER along with 99.5% and 98.6% recognition accuracy on ORL and Yale databases, respectively.

Generative Evolutionary Action Grammar Based Form Exploration

Abstract: The paper describes a prototype design system based on genetic algorithm to evolve concept designs, which are coded in an action grammar. The action grammar captures the design intents expressed by the designers’ strokes during sketching process. The model is based on the assumption that the strokes made by the designers embody the aesthetic intentions expressed through the product form. The knowledge about the design generation is captured and used by the artificial neural network and genetic algorithms respectively. Principal component analysis s is used for the tacit knowledge extraction in the form of heuristics. The proposed formalism is able to support the design within a family as well creative design. The model shows the promise of a conceptual design support system for the exploration of the form during early stages of the design.Copyright © 2013 by ASME

Finding Image Semantics from a Hierarchical Image Database Based on Adaptively Combined Visual Features

Abstract: Correlating image semantics with its low level features is a challenging task. Although, humans are adept in distinguishing object categories, both in visual as well as in semantic space, but to accomplish this computationally is yet to be fully explored. The learning based techniques do minimize the semantic gap, but unlimited possible categorization of objects in real world is a major challenge to these techniques. This work analyzes and utilizes the strength of a semantically categorized image database to assign semantics to query images. Semantics based categorization of images would result in image hierarchy. The algorithms proposed in this work exploit visual image descriptors and similarity measures in the context of a semantically categorized image database. A novel 'Branch Selection Algorithm' is developed for a highly categorized and dense image database, which drastically reduces the search space. The search space so obtained is further reduced by applying any one of the four proposed 'Pruning Algorithms'. Pruning algorithms maintain accuracy while reducing the search space. These algorithms use an adaptive combination of multiple visual features of an image database to find semantics of query images. Branch Selection Algorithm tested on a subset of 'ImageNet' database reduces search space by 75%. The best pruning algorithm further reduces this search space by 26% while maintaining 95% accuracy.