Adaptive Image Acquisition with Consideration of Camera Optics

Abstract

This dissertation addresses two common problems in image acquisition. We first introduce an adaptive image acquisition methodology by replacing the traditional birefringent filter with slight out-of-focus blur generated by the camera lens. The optimal defocus setting is automatically adapted to the power spectrum of the scene. A criterion to estimate reconstruction errors without the baseband knowledge of the scene is developed in this work. An optimal Wiener filter then recovers the captured scene and yields sharper images with reduced aliasing. The numerical and visual results for gray-scale images show that our method is superior to current acquisition methods.

The extension of the defocusing method to color image acquisition involves an extra demosaicking step. By designing a multichannel Wiener filter on the luminance and chrominance domain, we simplified the reconstruction of this problem. The error criterion defined for color acquisition is also improved on searching the optimal defocusing settings for the input scene. The acquired color filter array (CFA) image with the optimal amount of blur is reconstructed by a joint deblurring and demosaicking method. The simulation results show the defocusing acquisition achieves better image quality with fewer aliasing artifacts than the traditional acquisition method with or without an anti-aliasing filter.

An optimization of the spectral sensitivities of the Bayer CFA pattern is the other area we propose to address. Due to the nature of the optical sensor used in cameras, a CFA pattern is placed over the sensor to distinguish light spectra with different wavelengths. A multichannel Wiener filter is selected to determine the optimal sensitivity function for each color channel. We further optimize the green sensitivities for different noise levels. Simulation results show that the CFA with optimal spectral sensitivity functions delivers images with smaller color difference and better visual quality than the CFA with fixed sensitivities.