Laserfiche WebLink
MEMO 93 <br />2.3 Scan Line Errors <br />Landsat MSS imagery occasionally contains errors of striping and/or bad scan lines. The scan line <br />anomaly of striping is a linear error of incorrect pixel values along every seventh scan line horizontal <br />within an image. Striping of Landsat MSS data is commonly mitigated with a destripe algorithm (Catts <br />and Kohorram, 1985). ERDAS Imagine 9.0 contains a "Destripe" model that applies three vertical and <br />horizontal convolution kernels to an image, which minimizes the effects of striping (AppendixA.l). <br />A bad scan line contains saturated pixels in a single band or multiple bands of an image in one or more <br />adjacent scan rows (lines) horizontally across some or all of a scene. The ERDAS Imagine utility <br />"Replace Bad Lines" copies or averages a user supplied list of lines with the lines above and/or below the <br />affected lines (Appendix A.2). Table 1 quantifies the scenes acquired for SPDSS historic mapping that <br />are affected by striping and bad scan lines with percentage-affected estimates. Figure 3 is an example of <br />Landsat MSS imagery affected by striping and bad scan lines, (a) before ERDAS corrections, and (b) <br />after corrections. The bad scan line utility did not perform well on images with multiple adjacent bad <br />scan lines. The bad scan lines utility would correct errors in portions of the image while multiplying the <br />errors in other areas of an image. The ERDAS scan line correction utilities only mitigate the scan line <br />errors to a limited degree and were, therefore, applied with the aim of balancing user effort, which can be <br />considerable. The overall benefit was achieved. Bad scan lines had a noticeable impact upon the <br />preliminary classification as evidenced by rows of classified pixels with different crop types from the <br />majority of pixels within a parcel affected by bad scan lines. These effects were minimized in post <br />classification refinement procedures by attributing the majority crop type to each parcel. The effect of <br />striping and bad scan lines upon classification is discussed in the crop type classification, Section 4.2, of <br />this memo. <br />Fast Fourier Transformation (FFT) was assessed as an alternative method for correcting striping and bad <br />scan lines (Jensen, 1996). FFT is a technique for separating an image into its spatial frequency <br />components, when isolating repetitive anomalies, such as line errors, is theoretically possible. These <br />patterns would then be removed and the FFT image would be back-transformed. Upon analysis of a <br />back-transformed FFT image, it was discovered that the FFT editing process modifies all pixel values <br />within an image. This pixel modification had potential to alter irrigated lands classification and was <br />therefore abandoned. <br />Page 5 of 59 tij,Rareradde ~eshn~P~gy, ~os~ <br />