The image foresting transform (IFT) is a general tool for the design of image processing operators based on dynamic programming. Silicon image forest transform (SIFT) is a fast 8-bit data architecture for IFT-based operators in FPGA. It can implement queue-based methods such as morphological reconstructions, watershed transforms, shape saliences, distance transforms, skeletonization, edge tracking, with runtime gains from hundreds to thousands over the respective implementations in software. In this paper, we further extend the SIFT architecture to support 8- and 16-bit data and kernel-based operators such as dilation/erosion, smoothing, border enhancement, and sharpness filters. Moreover, we considerably improved the clock operation frequency and processing parallelism of SIFT.