Research On Camera Dust Spot Detection Based On Background Modeling And Target Modeling Methods

Recently, as digital products with camera function such as smartphone and tablet PCgeting more and more popular, the demand for mobile phone camera module gets popular aswell. Such camera module’s construction is simple, which usually only has a simple lens, anelectronic imaging sensor and relative circuit interface. In the actual production process,because the workshop is not an ideal environment without any dust, the imaging sensorcontacts air before the installation of lens, Froeign matter such as dust or scurf will easily getattach to the surface of imagimg sensor, result in the dust spot on the image producted by thefinal camera module, which has serious impact on the quality of imaging. In order to improvethe overall quality of the final products, it is demanded to detect dust spots for the cameramodules. However, as production scale of camera module increases, detecting dust spotsmanually has high labor costs, and the accuracy of detection depends on the experience andmental state of the worker, automization of dust spot detection will play an important role inimproving efficiency.This dissertation will focus on mobile phone camera module and propose an automaticdust spot detection method based on machine vision, detailed work of this paper is as follows:Several currently used dust spot detection method are briefly described, and actual mobilephone camera module dust spot detection images are combined, this dissertation point out that,under the circumstance that dust spot appears fuzzy and camera has apparent vignetting, usingthese methods directly will cause high rate of false detection. After analysing the principle ofdust spot imaging on mobile phone camera modules and the principle of vignetting, thenmodeling these respectively, this dissertation proposes a possible dust spot detection methodbased on background modeling with Least Squares and a dust spot searching method based ondust spot modeling with advanced normalized cross correlation. Experimental methods thenused to compare the detection results of these methods to the one of currently methods toanalyse the advantage of modeling methods on the actual camera dust spot detecion. Finally, acamera dust spot detection method which combines the above ones is proposed, and relatedexperimental data is shown later.Experimental results show that camera dust spot detection based on backgroundmodeling and target modeling methods has higher rate of detecting accuracy and lower rate offalse detection than the current detection methods under the circumstance that vignetting isapparent and dust spot is very fuzzy in the detecting image. The method proposed by thisdissertation has high practical value on automatic mobile phone camera module dust spot detection.