Image feature detection and matching in underwater conditions

Abstract

The main challenge in underwater imaging and image analysis is to overcome the effects of blurring due to the strong scattering of light by the water and its constituents. This blurring adds complexity to already challenging problems like object detection and localization. The current state-of-the-art approaches for object detection and localization normally involve two components: (a) a feature detector that extracts a set of feature points from an image, and (b) a feature matching algorithm that tries to match the feature points detected from a target image to a set of template features corresponding to the object of interest. A successful feature matching indicates that the target image also contains the object of interest. For underwater images, the target image is taken in underwater conditions while the template features are usually extracted from one or more training images that are taken out-of-water or in different underwater conditions. In addition, the objects in the target image and the training images may show different poses, including rotation, scaling, translation transformations, and perspective changes. In this paper we investigate the effects of various underwater point spread functions on the detection of image features using many different feature detectors, and how these functions affect the capability of these features when they are used for matching and object detection. This research provides insight to further develop robust feature detectors and matching algorithms that are suitable for detecting and localizing objects from underwater images.