Abstract: Since 1 of the 20 detectors in Terra MODIS band 5 (1.230-1.250 m) are noisy, there are sharp and repetitive stripe noise over the entire image. As for MODIS geolocated data, the stripe noise are irregular and sometimes uncontinuous, it brings a difficult problem to the image retrieving process. A detection method was presented to extract the stripe noise, and a maximum a posteriori (MAP) based algorithm was applied to correct the contaminated pixels. A local gradient based method was used to detect the abnormal pixels. In the MAP method, the likelihood probability density function (PDF) was proposed based on a linear image noise model, and a Huber-Markov model was employed as the prior PDF. The gradient descent optimization method was used to receive the destriped image. The proposed algorithm had been tested using a Terra MODIS band 5 geolocated image. The recovered images demonstrate that the proposed algorithm can remove the irregular stripes effectively. The power spectrum also shows a satisfactory result.