Max_missing, the convolution result is set to nan, signaling "not havingĮnough information here". Proportion of missing values within an overlap neighborhood exceeds We need to provide the user withĪ max_missing input argument to control this behavior: if the Proportion of missing is tolerable for the computation at location Squares, and they constitute 20% of the kernel size. In the interior case ( I), 5 missing values are represented by red The edge case ( E) covers at most 15 data points out of 25. For instance, theĬorner case ( C) in Figure 2a covers only 9 valid data points (at Take this edge-interior difference into account. Proportion of missing values in the neighborhood, we would like to At the edge or corner of the data domain ( Figure 2a,įewer overlapping values than in the interior. ( i, j) of the data domain the kernel defines a neighborhood within which theĬorresponding values in the kernel and the data are multiplied and the
R4 = (data, kernel, mode='same')Īs we slide the convolution kernel across the data, at each location The existing information contained in the data very well.įigure 1 also shows the results using other convolution functions inįigure 1c – 1e are computed using these functions: r2 = 2d(data, kernel, mode='same') For such kind of forgiving tolerance, the ()įunction would result in a big loss of data, and doesn’t utilize Least 50% of the neighborhood data to compute a neighborhoodĪverage. Maybe someĭegrees of data loss is tolerable, for instance, you require only at However, in other cases this might be too strict a requirement. Local neighborhood to compute the neighborhood mean. This is requiring that all data points are present and valid in a In some cases this might be the desired result: it is required thatġ00 % of the data in each convolution step need to be present and Inflated in the convolution result, because of the overlaps between It can be seen that the missing holes in the original data get (g) similar as (b) but using a custom convolve2DFFT() function. (f) similar as (b) but using a custom convolve2D() function. (b) Convolution of data in (a) with a 5×5 kernel (all 1s) using (). The missing values are drawn as white color. The kernel being convolved will create a missing in the output. That is to say, any overlap between the missing value and The convolved result will have an enlarged hole around each missingĭata point. These methods: If your data contain any missing value (e.g. Gradient computation/edge detection or the computation of Laplacian 2D convolution can be used to perform moving average/smoothing,
0 kommentar(er)
