oa An Interpolation-Based Stokes Image Reconstruction Scheme for DoFP Polarization Image Sensors

Summary

In this paper, we present a novel interpolation-based Stokes image reconstruction scheme for the division-of-focal-plane (DoFP) polarization image sensors. Different from the previous implementations, our proposed method first demosaics the raw image by mainstream interpolation algorithms then converts the up-sampled images to Stokes images with much richer polarization-related physical information. This not only leads to significant resolution improvement to the captured raw image, but also greatly reduces the caused pixel mean square error (MSE). Experimental data from the test images have validated the effectiveness of this proposed scheme.

Motivation

Solid-state image sensors, which are capable of extracting the incident light's polarization information in addition to intensity and color (i.e. wavelength), take great advantages in a wide range of applications [1]. By looking through a layer of patterned micrometer-scale pixelated polarizing elements, a set of mosaicked polarization raw sub-images can be generated simultaneously, namely DoFP polarization imaging. As shown in Fig. 1 (a), similar to the widely-exploited Bayer pattern of color imaging, the mosaicked polarization raw sub-images down-sample each polarization channel by 75%, leading to significant spatial resolution loss. Meanwhile, in order to make the raw sub-images physically meaningful, they are typically translated to Stokes sub-images, which correspond to first three Stokes parameters representing the unpolarized and linearly polarized components of the incident light. In the previously reported implementations, the neighboring four sub-pixels (i.e. I0, I90, I45, I135) are directly substituted to the Stokes parameters' classical expressions. As a result, a 75% down-sampled Stokes images are acquired. In order to compensate this resolution loss, we propose a new Stokes image reconstruction scheme: before the Stokes image conversion, interpolate the mosaicked polarization raw sub-images with mainstream algorithms, including linear, cubic and spline.

Results

Figure 1 (c) illustrates the proposed detailed image reconstruction flow. In addition, we have also compared our proposed method to the traditional implementation of directly applying the same interpolation algorithm to the aforesaid 75% down-sampled Stokes images [Fig. 1 (b)]. For the extracted down-sampled Stokes images in the previous implementations, even with the same interpolation algorithms applied, our proposed method still outperforms by almost 10 folds in term of Stokes parameter S2's MSE (Fig. 2), which well-balances the spatial resolution compensation and the Stokes image generation by minimizing the caused overall MSE. Figure 3 illustrates the real images with different Stokes image reconstruction schemes applied.

Acknowledgments

This work was supported by the National Natural Science Foundation of China (Grant No. 61504087), the Kongque Technology Innovation Foundation of Shenzhen (Grant No. KQCX20120807153227588), the Fundamental Research Foundation of Shenzhen (Grant No. JCYJ20140418095735624, and JCYJ20150324141711677).

Reference

[1] M. Kulkarni, V. Gruev, “Integrated spectral-polarization imaging sensor with aluminum nanowire polarization filters”, Optics Express, vol. 20, no. 21, pp. 22997–23012, 2012.