BID (Bruteforce Isoluminants Decrease) a RGB-to-gray conversion technique for automatic photogrammetry,

In the automatic reconstruction of three-dimensional objects and environments from two or more photographic sets, few solutions are able to take advantage of color information. Many of these reconstruction methods are conceptually designed to work on grayscale images in the sense that, sooner or later in the processing, for a given spatial location, the algorithm will only consider a single intensity value instead of the RGB triple. Color to grayscale conversion is factually a dimensionality reduction problem. This process should not be underestimated, since there are many different properties that need to be preserved. Isoluminant color changes are usually not preserved with commonly used color to gray conversions. In any case, we can state that the 3D to 1D dimension reduction leads to information loss and that the appearance of this loss is related to the method. Many conversion methods have been proposed in recent years; these methods mainly focus on perceptual accuracy in terms of the fidelity of the converted image when reproduced from color to grayscale tones. These kinds of approaches are not designed to fulfill the needs of visual stereo matching and image matching algorithms, where local contrast preservation is crucial in the process of matching by local operators. This is one the main reason why in Lowe’s “Scale-invariant feature transform” (SIFT) operator, the candidate keypoints with low contrast are rejected in order to decrease the number of ambiguous points in the matching process [1]. To face this issue, we evaluated some State-of-the-art algorithms designed to perform the so-called RGB-to-gray conversion. The results of this analysis phase led to develop a new procedure in which the most promising algorithm, instead of evaluating every single image separately, has been adapted to evaluate and find the best possible converting solution among the entire set of images. In this paper, we present the BID (Bruteforce Isoluminants Decrease) a RGB-to-gray conversion technique that combines the idea of Multi-Image Decolorize (MID) [2] with our specifically developed framework. More specifically, the MID evaluates a whole set of images instead of a single one in order to preserve tonal coherence during the matching phase. On the other hand, our frameworks specifies the statistical properties of the input data with the help of a representative collection of image patches provided by the same images of which we realize the conversion. Differently from MID, that is an adaptation of the Grundland & Dogdson algorithm [3], our conversion is a generalization of the MATLAB RGB2Gray algorithm, and simultaneously takes as input the whole set of images to be matched. BID bears some similarity to the previously introduced by Song[4]. However, significant critical difference lies in the measurement criterion used to evaluate the decolorization quality. In brief, Song et al. employs the bilateral filtering with high computational complexity; on the contrary, BID is based on the newly defined dominant color hypothesis and aims to maximize the tonal representation on the image set.