Class-Driven Color Transformation for Semantic Labeling
نویسندگان
چکیده
We propose a novel class-driven color transformation aimed at semantic labeling. In contrast with other approaches elsewhere in the literature, our approach is a supervised one employing class information to learn a color transformation. Our method maps image color to a target space with maximum pairwise distances between classes and minimum scattering within each of them. To compute the color transformation, we pose the problem in terms of a composition of two mappings. The first mapping employs a pairwise discriminant cost function minimized through a steepest descent optimization to map the image color data onto a space spanned by the class set. It targets better separability between distinct classes as well as less scattering within each individual class. The second mapping corresponds to subspace projection of this class data to a target space with same dimensionality of image color data. To preserve distances attained by the first of the mappings, this subspace projection is effected making use of metric multi-dimensional scaling. We report our experiments on MSRC-21 and SBD datasets, where our method consistently improves overall and average performances of well-known publicly available TextonBoost and DARWIN multiclass segmentation frameworks at a negligible computational cost. These results confirms our contribution towards reflection of higher distinction in color space by imposing better separability in a novel representation which is learned from class information of the dataset under consideration.
منابع مشابه
VHR Semantic Labeling by Random Forest Classification and Fusion of Spectral and Spatial Features on Google Earth Engine
Semantic labeling is an active field in remote sensing applications. Although handling high detailed objects in Very High Resolution (VHR) optical image and VHR Digital Surface Model (DSM) is a challenging task, it can improve the accuracy of semantic labeling methods. In this paper, a semantic labeling method is proposed by fusion of optical and normalized DSM data. Spectral and spatial featur...
متن کاملModel Transformation for Model Driven Development of Semantic Web Enabled Multi-Agent Systems
Semantic Web evolution brought a new vision into agent research. The interpretation of this second generation web will be realized by autonomous computational entities, called agents, to handle the semantic content on behalf of their human users. Surely, Semantic Web environment has specific architectural entities and a different semantic which must be considered to model a Multi-agent System (...
متن کاملبرچسبزنی نقش معنایی جملات فارسی با رویکرد یادگیری مبتنی بر حافظه
Abstract Extracting semantic roles is one of the major steps in representing text meaning. It refers to finding the semantic relations between a predicate and syntactic constituents in a sentence. In this paper we present a semantic role labeling system for Persian, using memory-based learning model and standard features. Our proposed system implements a two-phase architecture to first identify...
متن کاملبرچسبزنی خودکار نقشهای معنایی در جملات فارسی به کمک درختهای وابستگی
Automatic identification of words with semantic roles (such as Agent, Patient, Source, etc.) in sentences and attaching correct semantic roles to them, may lead to improvement in many natural language processing tasks including information extraction, question answering, text summarization and machine translation. Semantic role labeling systems usually take advantage of syntactic parsing and th...
متن کاملAspect Oriented UML to ECORE Model Transformation
With the emerging concept of model transformation, information can be extracted from one or more source models to produce the target models. The conversion of these models can be done automatically with specific transformation languages. This conversion requires mapping between both models with the help of dynamic hash tables. Hash tables store reference links between the elements of the source...
متن کامل