/rrsc-mars

Primary LanguageJupyter NotebookMIT LicenseMIT

Crater and Sand Dune detection on Mars and Moon

  1. Preprocessing (Both Optical & DEM)

    1. Noise Filter The Median filter will be used with a 5X5 window size.
    2. Area Filter The filter is applied on the Median Filtered and DEM image using OTSU's method. Features with less than 1000 px are removed
    3. Shape Filter Disk shape with radius 5px is used. The area filtered binary image and DEM is converted back to grayscale using matlab function im2uint16. The shape filter is applied on the resulting images.

  2. Crater Detection

    1. Optical Image

      1. Wavelet Transform Wavelet transform has been applied on the preprocessed image (TMC and MiniSAR) to get multiscale 2-D wavelet decomposition of gray scale preprocessed image and the corresponding detailed coefficients (horizontal, vertical and diagonal edges responses). The steps:

        1. Read Original Image
        2. Apply Preprocessing
        3. Multiscale 2-D Wavelet (Haar) decomposition of gray scale image has been done. The Haar wavelet applies a pair of low pass and high pass filters to image decomposition. Upto five levels (scales) of decomposition have been done to get the detailed edges of craters for TMC image and three levels for MiniSAR images. The inbuilt function ‘wavedec2’ of MATLAB has been applied on the image for all the five levels and it returns corresponding wavelet decomposition structure having the decomposition vector C and the corresponding matrix S for all the five levels. The size of vector C and the size of matrix S depend on the type of analyzed image. In our case, C is having (3n+1) sections and size of S is (n+2) - by – 2, where n is the level of decomposition.
        4. After decomposition of the image at five levels, it is reconstructed for corresponding levels.
        5. The corresponding detailed coefficients (horizontal, vertical and diagonal edge responses) for all the five levels have been achieved by passing the wavelet decomposition structure, obtained in previous step, to a MATLAB inbuilt function ‘detcoef2’ as input parameters.
        6. Visualize all the detailed coefficients for each level as an image.
        7. Continuous Wavelet Transform using Haar wavelet has been applied on the image with a scale (s) varying from 1 to 32 to get image details in the form of continuous wavelet coefficients using MATLAB inbuilt function ‘cwt’.

      2. Generalized Hough Transform Hough transform is generally used for edge linking and boundary detection purposes in the image segmentation process. Results of edge detection methods may contain sparse points, instead of straight lines or curves. Therefore, there is need to fit a line to these edge points. In the methodology of this work, generalized Hough transform has been used as an image segmentation technique in image based crater detection approach applied on TMC and MiniSAR images. The generalized Hough Transform has been implemented in MATLAB, as per the given equations (Ballard, 1981) for the input preprocessed image. Steps :

        1. Read the original image.
        2. Preprocessing of the image is done.
        3. The Sobel edge detector has been used to get all the edges of objects (craters) in the preprocessed image. Hence binary egde image is obtained.
        4. Horizontal and vertical edge components (GH and GV) are obtained from the Sobel edge detector.
        5. Finally for each detected edge pixel, the shortest distance, r, (Euclidean distance) from edge point (GH, GV) to the reference point (Xc, Yc) has been calculated as per the equation 4.1. Initially ) are coordinates of origin (0, 0).
        6. Direction, β, (angle) has been calculated using GH and GV as per the equation 4.2. It is the inverse tangent angle of the ratio of GV to GH. This direction is also called aspect.
        7. Update the values (Xc, Yc, r, β) for each value of distance and angle as per the equations 4.3, 4.4, 4.1 and 4.2 respectively.