Peningkatan Ekstraksi Fitur Berbasis Scale Invariant Feature Transform Menggunakan Metode Multiscale Retinex Untuk Meningkatkan Jumlah Keypoint
The purpose of this research is to perform the improvement of feature extraction based on SIFT algorithm with MSR preprocessing step to increase the number of keypoint. Often in recognizing the SIFT algorithm pattern is affected by image quality. The better the recognizable image, the more number of keypoints it can get. For that need to do preprocessing stages so that the accuracy produced SIFT higher. In this study the researchers tried to combine SIFT algorithm with Multi-Scale Retinex (MSR) method. The MSR method is used because it has advantages in improving image quality in images that have less illumination (darkness). From the test results using MSE and PSNR the proposed method (MSR) is better than the previous methods. The average test result of number of keypoint without preprocessing is 1083.523, Constrast Stretching is 1093.797, CLAHE is 1105.891 and MSR is 1399.162. From this test it can be concluded that the SIFT combination with the preprocessing stages of Multi-Scale Retinex produces a good number of keypoints compared with other methods and without preprocessing stages.