روشي جديد براي افزايش دقت مكاني تصاوير سنجش از دور با استفاده از جدول جستجو

A novel method for increasing the spatial resolution of remote sensing images using lookup table

تحقيقات مختلفي براي افزايش دقت مكاني تصاوير با هدف تجزيه پيكسل ­هاي مخلوط در سنجش از دور انجام شده است. اين روش­ها و مشكلات پيش­رو در اين تحقيق بررسي خواهند شد. در ادامه مبحث جديدي براي افزايش دقت مكاني تصاوير پيشنهاد خواهد شد. در اين روش يك تصوير با دقت مكاني كمتر، از تصوير ورودي استخراج مي­شود كه همانند يك جدول جستجو عمل مي­كند. با تعريف يك معيار براي شباهت بهينه پيكسل ­ها در دو تصوير، براي هر پيكسل از تصوير ورودي يك پيكسل مشابه در جدول جستجو مي­يابيم. نشان خواهيم داد كه پيكسل­ هاي مشابه در دو تصوير، داراي يك ساختار مشابه از زيرپيكسل ­هاي تشكيل­دهنده آن پيكسل مي­ باشند. با استفاده از اين روش هر پيكسل مخلوط به تعدادي زيرپيكسل كه اغلب خالص هستند، تجزيه خواهد شد. شبيه­سازي­هاي انجام شده بر روي اطلاعات ساختگي و واقعي نشان­ دهنده بهبود شاخص هاي طبقه ­بندي توسط روش پيشنهادي و تجزيه بهتر پيكسل­ هاي مخلوط، نسبت به روش ­هاي قابل مقايسه است.

Different methods have been proposed to increase the image spatial resolution by mixed pixels decomposition. These methods can be divided into two groups. Some research have been attempted to obtain percentages of sub pixels and the other try to obtain their locations. These methods and their problems will be examined in this study. Common methods are reviewed with more emphasis. Finally, a new method for increasing the spatial resolution will be proposed to resolve some deficiencies of existing methods. Especially this method, instantly takes percentages and locations of mixed pixels end members without no use of additional information. This method applies a proper lookup table, which is derived from an input image. By defining a similarity metric function, we obtain a similar pixel for every input pixel. These similar pixels have equal sub pixel structures; hence, an input pixel will be decomposed to a proper set of sub pixels. In the high quality images, these sub pixels usually, belong to pure classes. This proposed method is examined on four sets of artificial and real data. First we degrade these data sets by averaging filtering, and then we restore degraded data, using this method and two other methods. One of these methods is a hard classification and the other is a combination of fuzzy c-means and direct method to obtain percentages and locations of sub pixels respectively. We obtain percent of correction classification and KAPPA criterions for these methods. Simulation results on artificial, real data show a good sub pixels decomposition performance of proposed method relative to those of other comparable methods. By particular, this method shows at least 7% of improvement in artificial and 2% in real data relative to other methods.