Block Based Focus Volume Pyramid for Shape from Focus

In shape from focus (SFF), the quality of reconstructed depth map depends on the focus volume (FV) obtained from the image sequence. An FV contains detailed focus information for every pixel in each image of the image sequence and is obtained when a focus measure Operator (FM) is applied on the image sequence. A common issue within FMs is their sensitivity to noise, which can significantly compromise the accuracy of the depth extracted from the FV obtained. To address this, we propose reconstruction of the FV by using a block-based approach. Our approach involves block processing of the FV and then using a statistical measure, the trimean value, for each block to effectively suppress noise. Since the trimean value is independent from the extreme values of the data, it helps in removing the noise. This process is carried out for multiple block sizes, after which we utilize a pyramid approach to aggregate the reconstructed FVs from the varying block sizes, ensuring that no important information is lost from the image sequence. Experiments conducted on both synthetic and real image datasets have proven that our method produces accurate depth maps that are robust to noise.