We study the influence of edge areas on hyperspectral subpixel target detection. In this paper, two methods having the potential to overcome detection difficulties at edge areas are presented. The first method proposes a new directional filter which is a derivative of the standard anomaly match filter. When applying the directional filter, we estimate the test pixel's background assuming that the central pixel should be similar to the neighboring surrounding pixels. However, simply taking the mean of the surrounding pixels will lead to errors at edge areas. Instead, we take the two most appropriate adjacent signatures suspected of having the same background characteristics as the pixel being tested. This way we eliminate the strong dependency existing between the degree of edge content and the false alarms calculated using the local mean filter. Another method developed is based on applying a linear PCA (Principal Components Analysis) transformation on the image and using a local mean match filter algorithm while attenuating the first PCA channels. This way, that part of the signature orthogonal to the sub-space being spanned by those channels dominated by the background will play a more significant role than before. The primary sources of these components have to be target signatures or stationary noise, so they are less influenced by the pixel location. A number of correlation techniques, developed to check the connection between the anomaly vector and the "edginess" values in the datacube, are shown. The results indicate that there is a strong dependency when using the local mean filter in most images, while it is lowered drastically after applying the PCA normalized match filter algorithm and, in some cases, the directional filter.