It is possible to achieve improved color accuracy with a color camera by placing a color filter in front of the camera. Unfortunately, the color filter will block some of the light entering the camera, which will result in additional noise in the recorded data. This paper provides an initial investigation into finding an optimal solution to the filter design, in the presence of noise.
The properties of prints are not fully determined by the materials they are composed of and the method that was used to compose them. These merely set limits to what a print's properties, such as its colors, sharpness, smoothness, color inconstancy and level of ink use, will be and it is the role of a printing system's imaging pipeline to select a particular combination. Conventionally such choices are implicit in how resources for a pipeline are built and can be improved with experience and trial an error. Performance can be improved though by optimizing for specific attributes, as was previously shown for color consistency, ink use and grain among others. A key constraint that remains here is that optimization is performed on the basis of sampling and search strategies, which have inherent limitations. This paper presents a direct, analytical approach to optimization that hinges on the insight that it can be performed in a convex space even when the properties involved in the optimization do not relate to each other in a convex way. The result both improves performance versus previous methods and does so in considerably less time.
Mutual Information (MI) is emerging as a very strong metric for image registration purposes in the literature. It has many applications from remote sensing to medical image registration. From this wide range of use of MI, images are mostly expressed in different numbers of bits (high dynamic range) especially in medical and satellite imaging. In such cases, contrast enhancement becomes inevitable before MI-based image registration since all the images should be in the same intensity range. The change in intensities in images will directly affect MI metric. Contrast enhancement methods also have a significant effect on the registration performance due to MI metric and this problem is not sufficiently addressed in the literature. In this paper, the effect of the outstanding contrast enhancement methods is examined on image registration performance. For this purpose, high dynamic range satellite images were used and Monte Carlo tests were performed. They are tried to be aligned with MI and constrained optimization by linear approximations (COBYLA) optimization algorithm. Consequently, it is found that contrast enhancement methods have an effect on MI-based image registration. It is concluded that Laplacian of Gaussian unsharp blending masks (LoGUnsarp), adaptive histogram equalization (AHE) and contrast limited adaptive histogram equalization (CLAHE) methods have better registration performance. They can be preferred in such registration purposes.
The paper presents an efficient algorithm that reduces the time complexity of video coding in the H.265/HEVC encoder, towards an implementation employable in real-time video coding and transmission applications. The optimization targets the motion estimation search procedure, which occupies a large part of the compute time per Coding Unit. Experimental results demonstrate extensive processing time savings while maintaining similar compression quality and bit rate as the standard.