Abstract. The tasks of image reconstruction from measured data and the analysis of the resulting images are more or less strictly separated. One group of scientists computes by applying reconstruction algorithms to the images; the other then operates on these images to enhance the analysis. First attempts at combining image reconstruction and image analysis, in a nonsystematic way, are known as Lambda tomography or Tikhonov-Phillips methods with 1-norms or with level-set methods. The aim of this paper is to provide a general tool to combine these two steps; i.e., even in the reconstruction step the future image analysis step is taken into account, leading to a new reconstruction kernel. Here we concentrate on linear methods. As a practical example we consider the image reconstruction problem in computerized tomography followed by an edge detection. We calculate a new reconstruction kernel and present results from simulations.Key words. image analysis, image reconstruction, approximate inverse
AMS subject classifications. 65R32, 45Q05DOI. 10.1137/0707008631. Introduction. In order to extract information from a given image, analysis tools are used. In a first step one applies operators on that image; then the searched-for information is found by operating on these enhanced versions of the original picture. Images typically are twodimensional arrays of numbers. Of course, three-dimensional arrays for volume data or even time-dependent data, which may amount to four-dimensional data, are conceivable. Prominent analysis tools are edge detection methods where first partial derivatives of smoothed versions of the image are computed, followed by recognition methods. A typical example is the Canny edge detector; see [4]. Other operations can be found, e.g., in [6,11]. Here we restrict our discussion, as mentioned above, to linear operators. In denoising one can think of solving the heat equation with homogeneous boundary conditions and the original image as initial condition, at the final time T the image is considered to be denoised.We have to mention, of course, that nonlinear methods also play an essential role. But this does not-at least at the moment-fit into our framework.Attempts to combine reconstruction and analysis are known, but are not systematically pursued. As an example we mention the Λ computerized tomography, in which local inversion formulas produce images where the singular support is preserved; this means that those images have jumps wherever the original image has them (see, e.g., [10,13,21]). The use of TikhonovPhillips regularization with 1 -norms results in smooth images; see, e.g., [5]. Level-set methods