The Semantic Gap Analysis

1894 Words 8 Pages
Overview of the Semantic Gap - What is it? The semantic gap is the difference between human perception of observations, activities, and objects and their computational or machine-based representations(). It is a split between high level features, or semantic information, and low-level features, of which there are many types. High-level features can include keywords, concepts, categories, or ontologies - virtually all things that lead to determining meanings of text-based information and phrases. Low-level features can include color, texture, resolution, encoding, salient points, etc. which are all measurable aspects of content-based information, or non-text information, such as images and videos. The spread of multimedia on the web is widening …show more content…
(2) What is latent semantic indexing, and how has it impacted the research of CBIR and bridging the semantic gap? (3) How are 5 current methods impacting CBIR differently, and which are best for bridging the semantic gap? (4) What limitations have there been in the current research, and why are they worth investigating? (5) What are some future implications for the semantic gap, and how to better approach the research process for CBIR? The layout of the paper will present the rationale of why the semantic gap is an important area concerning IR, discuss background work pertaining to the semantic gap and image retrieval, and CBIR with high level semantics. What has been learned in the course will then be discussed, as well as some insights gained from what was learned. A personal assessment will follow, highlighting areas of research that need to be further investigated, such as balancing methods and emphasizing user experience research more. Finally, conclusions of progressions made in the research as well as areas for future study will be followed by a closing section on what the reader should have learned from the topics …show more content…
It is the simplest method and an ideal choice for aiding in the semantic gap challenge, since color can be treated as a global feature, and color histograms are organized for easy comparison. While human perception is poorer at analyzing histograms, this technique is most ideal for machine learning, since color constancy, salient points, and texture can be interpreted by a computer. Individually, these low-level features do not always yield the best results, but combined, the results improve. Using feature models to combine color constancy and spatial frequency help the machine to interpret distinct points in an image. In some cases, this improves search for overall images, but for close ups, there may be too few features for the machine to properly represent an accurate result

Related Documents