Abstract: An approach of localizing a mobile robot with compressed omnidirectional images for robot localization and environment modeling is presented. The Parzen density estimation algorithm in omnidirectional image compression is discussed. Fast Fourier transform helps in lowering the computing complexity of Parzen density estimation and in attaining real time processing. Principal components analysis algorithm is used to reduce further the dimensionality of feature vectors. A fast algorithm called singular value decomposition is used to get the first k principal components of the set of feature vectors. Localization is performed by projecting image data of the robot's current position into the eigenspace forming a point and finding the closest neighbor point to previously acquired images on the environment modeling. The proposed method is tested through many experiments undertaken in indoor environments.