Filtering technique for image analysis

Filtering technique for image analysis

Maggi, F.

Civil Engineering and Geosciences

Hydraulic Engineering

2003-06-13

The measuring of the flocculation parameters is based on the analysis of the optical records obtained with the settling column. These records consist of images of aggregates or flocs grabbed with a digital camera. Flocs are lightened by means of a laser sheet coming from the side. The laser light is scattered by the flocs and recorded by the camera in digital format. The records result in intensity images that need to be elaborated in order to separate the object of interest from the surrounding environment. This report deals with the aforementioned processing technique, thus describing which solutions have been chosen and which treatments have been developed. Filtering techniques for image separation (or segmentation) and noise reduction in image analysis have become a discipline on itself, due to the number of algorithms that have been developed in the early decades. In this report we will not describe all those techniques, but we rather develop ourself a simple separation technique which better suites the purpose of our investigation. Therefore, we first describe the domain of our records and their characteristics, then we present the separation technique which we test on a set of images generated artificially. Subsequently, we consider noise reduction techniques in combination with it and we select an optimal combination of filters by comparing error-based statistics. Finally, we infer the behaviour of the procedure when applied to images of different resolution, by developing an analytical description of the error as a function of the resolution. The most important issues highlighted in this document are± - We have identified a criterion to separate the foreground (floc) from the background (water medium) based on the pdf intensity distribution. - Tests were carried out which resulted in an optimal combination of filters which consists of the median filter and the pdf-based seperation technique. - We have observed that the filters remove large parts of noise within the floc mass but concentrates the error at the boundary. - We have concluded that the error at the boundary of the flocs does not affect the overall detection of the perimeter-segmentation. - The processing technique is sensitive to the noise wideness by a second order effect. - The filtered images are affected by an error of about 0.02 for large flocs and 0.2 for small flocs by estimation.

filtering technique
image analysis
filtering
flocs
optical
noise

http://resolver.tudelft.nl/uuid:4e33754c-ba9c-4fe9-84d0-3e6a69eac207

TU Delft, Department of Hydraulic Engineering

Report no. 2-03

Institutional Repository

report

(c) 2003 TU Delft, Department of Hydraulic Engineering