Your best bet would be to mosaic the raw red band and near infrared band images from which the NDVI images are derived. There are techniques for creating seamless mosaics for images, e.g. through the use of histogram matching and feathering techniques. For areas of overlap, the feathering method will calculate the output value as a weighted combination of the two input pixel values from the overlapping bands, where the weights are derived from the squared distance of the pixel to the edge of the data in each of the input raster files. Therefore, less weight is assigned to an image's pixel value where the pixel is very near the edge of the image. The result of this feathering method is that the output mosaic image should have very little evidence of the original image edges within the overlapping area. Histogram matching can be performed on one of the input images (i.e. the 'Append' image) to force its radiometric properties (contrast) to match that of the 'Base' image. The best results will be from combining histogram matching and feathering.
Once you're satisfied with the mosaics, you can then derive the larger NDVI image (NDVI = (NIR - Red) / (NIR + Red)). This will likely result in a more satisfactory mosaic than trying to match the NDVI images directly.