Remote sensing images are vulnerable to many degradation sources. According to the characteristics and complexity of the image degradation, the issue of information enhancement can be classified into three levels, as shown below: 1) information restoration; 2) non-uniform information correction; and 3) missing information reconstruction. The SENDIMAGE laboratory is devoted to undertaking comprehensive and systematic research in this field. Numerous achievements have been published in international journals, including more than 10 papers in IEEE TGRS. Furthermore, the research results have been applied to the pre-processing systems of the domestic satellite data.

The framework of remote sensing information enhancement

1.Information Restoration (Denoising & Deblurring)

   Information restoration aims at recovering the original clean image from a noisy or blurry image. According to the different degradation characteristics, niche targeting noise reduction methods have been proposed in the SENDIMAGE laboratory, including the removal of stripe noise, spectral noise, temporal noise, polarization noise, impulse noise, and multiplicative (speckle) noise. Meanwhile, the restoration problems of sensor defocus and atmospheric haze have also been studied in depth in the laboratory, and we have proposed a novel high-fidelity method for the removal of atmospheric haze.

  • Strip Noise Removal

  • Spectral Noise Removal

  • Multiplicative (Speckle) Noise Removal

  • Polarizing Noise Removal

  • Impulse Noise Removal

  • Temporal Noise Removal

  • Defocus Image Restoration

  • Atmospheric Haze Removal

2.Correction of Nonuniform Information

   The correction of non-uniform information aims to correct non-uniform brightness distribution in images, which can be caused by a number of factors. The sensors, atmosphere, topography, surface features, and artifacts are considered to be the main factors. We have carried out research into the relative calibration of linear-array sensors, the non-uniform correction of area-array sensor images, thin cloud correction, topographic correction, building shadow correction, and seamline correction.

  • Non-uniform Information Correction

  • Thin Cloud Correction

  • Topographic Correction

  • Building Shadow Correction

  • Seamline Correction


3.Missing Information Reconstruction

   This technique aims to reconstruct the missing information caused by sensor failure and thick cloud cover, and is also applicable for object removal. The SENDIMAGE laboratory has undertaken comprehensive research into this topic, from the aspects of spatial-based reconstruction, temporal-based reconstruction, spectral-based reconstruction, and spatio-spectro-temporal combined reconstruction.

  • Spatial-based Reconstruction

   The spatial correlation is modeled for a single image to complete the missing information using the remaining information in the target image under the situation of no complementary information being available. The spatial-based methods can also be applied to information reconstruction, object removal, object disappearance, etc.


  • Temporal-based Reconstruction

   Temporal-based reconstruction is based on multi-temporal observation data. The missing information is reconstructed by the modeling and completion of the complementary information among the data. The following figure is an example of cloud removal for remote sensing imagery.

  • Spectral-based Reconstruction

   In multispectral and hyperspectral images, there is much redundant spectral information, due to the characteristics of the sensors. This redundant information can be used to reconstruct the missing data in a specific band. The classical application is the reconstruction of the missing information of Aqua MODIS band 6 using the complementary multi-spectral information. The error of our proposed method is within 5%.

  • Spatio-spectro-temporal Combined Reconstruction

    Making the best use of the complementary spatial, temporal, and spectral information, the missing information reconstruction is undertaken with a combined or integrated model. For the missing information of Landsat ETM+ data, the SENDIMAGE laboratory proposed a robust spatio-temporal reconstruction method and also developed a software tool, which can be freely downloaded from the laboratory website.

4.Engineering Application

   For domestic remote sensing satellite data, we have proposed a number of advanced information enhancement algorithms and developed practical software solutions, some of which have been running operationally. For the products, we obtained better results than those processed by the MDA ground pre-processing system provided by the famous geographical information technology company of Canada.