Robust Image Feature Description, Matching And Applications: A Review

Abstract

Many computer vision applications such as visual communication and image processing, object detection, shape recognition, recognition of face and face expression, 3D reconstruction, etc. included two images to suit. To compare two images, or in other words, to evaluate the similarity / difference between the two images, a certain picture definition is required since the comparison between the raw intensity values of two images takes more time and is influenced by small differences in the inherent properties of each image, such as luminosity, orientation, scaling etc. The photographs may then be corresponded with their definition extracted from the fundamental characteristics of the picture, such as colour, texture, structure, etc. The actual descriptor / signature of the picture is this definition. Any descriptor's main objectives are (1) to collect discriminatory picture details, (2) to provide the invariance to geometric and photometric modifications, and (3) to - its size. The key objective of the paper is to construct the image descriptors with differential strength, image variations robustness and small scale. For regionally based photos under different geometric and photometric transformation conditions we have provided an interlaced strength value local descriptor (IOLD). We have checked four local gray-scale image descriptors, namely Local Extremity Diagonal (LDEP), a Local Bit plane Diagonal (LBDP) pattern, local LBDISP and local wavelength (LWP) pattern in the MRI and CT repositories for biomedical image retrieval. Four colour-based local descriptors, i.e. local colour occurrence descriptors (LCOD), robust hybrid (RSHD) scale and rotating, multiple-channel adder based, and multi-channel decoder-based, local binary (md LBP) patterns for natural and texture image recovery, have been reported. For more details, see LCOD. As a favored phase in pre-processing an illumination compensation mechanism was recorded. Filter bags and SVD-based solutions were suggested in order to boost descriptor efficiency.