Low Cost Edge Sensing for High Quality Demosaicking

Abstract

Digital cameras that use Color Filter Arrays (CFA) entail a demosaicking procedure to form full RGB images. As today's camera users generally require images to be viewed instantly, demosaicking algorithms for real applications must be fast. Moreover, the associated cost should be lower than the cost saved by using CFA. For this purpose, we revisit the classical Hamilton-Adams (HA) algorithm, which outperforms many sophisticated techniques in both speed and accuracy. Based on a close look at HA's strength and weakness, we design a very low cost edge sensing scheme. Briefly, it guides demosaicking by a logistic functional of the difference between directional variations. We extensively compare our algorithm with 28 demosaicking algorithms by running their open source codes on benchmark datasets. Compared to methods of similar computational cost, our method achieves substantially higher accuracy; Whereas compared to methods of similar accuracy, our method has significantly lower cost. Moreover, on test images of currently popular resolution, the quality of our algorithm is comparable to top performers, whereas its speed is tens of times faster.

A Compact High-Quality Image Demosaicking Neural Network for Edge-Computing Devices

“…The reconstruction accuracy of the demosaicking method is important, but blindly pursuing high PSNR while ignoring the computational cost loses its meaning for practical application. As mentioned in [ 22 ], current works on demosaicking have achieved top demosaicking accuracy on benchmark datasets which is high enough. So running speed and memory required to store model parameters (especially for the CNN-based method) are the main issues that should be considered first.…”

“…The CNN-based method, which is a kind of data-driven approach, on the other hand, can automatically extract image features and complete the reconstruction process, making them suitable for implementation on some AI accelerators [15][16][17] or vision chips [18][19][20][21]. Although they improve the quality of the demosaicked image markedly, they need to store a large number of the model parameter and cost huge computational time [22]. Several deep learning-based demosaicking methods proposed in recent years focus too much on demosaicking accuracy while ignoring the computational cost of the algorithm.…”

A New Polarization Image Demosaicking Algorithm by Exploiting Inter-Channel Correlations With Guided Filtering

“…The entire denoised image q and the confidence level L can be generated by calculating i q and i L at each pixel from (10) and (15). For simplicity, this calculation can be expressed as a single function GF in (16), where the input arguments to the function are a target image p to be denoised and a guidance image I, and the outputs of the function are q and L .…”

“…It is used to reconstruct a high-quality full-resolution image from the down-sampled polarization sub-images. As each pixel element of the image sensor can detect only one channel of raw image data, the data of the missing channels will have to be predicted from the spatially shifted sensor data [16]. Since spatial-limited signals can never be band-limited, aliasing is inevitable when a full-resolution image is reconstructed by sampling the raw image data, resulting in zipper, fringes and other visual artifacts.…”

A Hybrid Denoising Algorithm of BM3D and KSVD for Gaussian Noise in DoFP Polarization Images

“…Giant strides have been made towards finding ways to remedy this limitation by interpolation algorithms dedicated to DoFP imagers [10]- [12]. Since each pixel of the DoFP image sensor can only record one component of the full polarization information, interpolation algorithms are used to recover the missing polarization information and reconstruct the full-resolution image from the down-sampled polarization sub-images [13]. The interpolation algorithms for DoFP imagers, also known as ''polarization demosaicing'' algorithms, serve a similar purpose to what the ''CFA interpolation algorithms'' do for imagers that are based on Bayer Color Filter Array.…”