REMOVING noise from images is an essential task since good quality images are required in various applications such as medical imaging, satellite imaging, recognition, etc. There are several types of noises which can degrade the quality of images. One such type is salt and pepper (SAP) noise. This noise can be represented as randomly occurring white (1) and black (0) pixels in the image. The main sources of this noise are electrical conditions, light intensity, imperfection in imaging sensors, transmission errors, etc. [1], [2]. The state-of-the-art suggests that various approaches have been proposed for removal of SAP noise. Median filter [3] and adaptive median filter [4] are popularly used for SAP noise removal. In these approaches, noisy pixel intensity is replaced by the median of intensities of neighborhood pixels. Although these filters are efficient to remove noise but fails to preserve details of the image due to blurring at the edges. The weighted mean, weighted fuzzy mean, adaptive fuzzy mean and iterative fuzzy filters are also used to reduce SAP noise [5]–[۸]. In all these methods, the problem lies with weighing of good pixels which may lead to loss of actual image details to a certain extent. Ahmed et al. [9] have proposed an iterative adaptive fuzzy filter for removal of high-density SAP noise. The drawback of this approach seems to be assignment of weight to good pixels in window during denoising using inverse distance weighting function. Therefore, this method fails to preserve the image details. The other problem of this method is use of many heuristic parameters such as , K1, and Department of Electrical Engineering, IIT Kanpur, India – 208016 K2 which are not consistent to give best result for different noise levels. To overcome these problems others have proposed fuzzy filter based on Type-2 fuzzy set [10]–[۱۲]. Liang and Mendel [13] proposed Type-2 adaptive filter using an unnormalized Type-2 Takagi Sugeno Kang (TSK) fuzzy logic system (FLS) for the application of equalization of a nonlinear time-varying channel. John et al. [14] have applied neuro-fuzzy clustering techniques for classifying images where an image is represented by Type-2 fuzzy set. Yıldırım et al. [15] have proposed a Type-2 fuzzy filter for suppressing noise in the image while at the same time preserving thin lines, edges, texture, and other useful features within the image. In [16], [17], Type-2 fuzzy filter for edge, corner detector and noise reduction from the color images have been proposed. The drawback of these methods are formation of big fuzzy rule base (FRB) matrices and use of rigorous fuzzification and defuzzification, which increase computation time and complexity of the filter. The filter window is also not adaptive with respect to noise level. In this paper, we propose an adaptive Type-2 fuzzy filter with a combination of Type-1 fuzzy set to eliminate the problem of FRB matrix formation, fuzzification and defuzzi- fication. The proposed filter consists of two steps. In the first step, pixels are categorized as good or bad. In the second step, the pixel categorized as bad in step-1 is denoised. For step1, two approaches based on adaptive threshold using primary MF values of Type-2 fuzzy logic are developed. Either of the two approaches may be used in the first step. In the second step, for denoising the bad pixels, good pixels are weighted in their respective filter window. To assign proper weight to good pixel a novel Type-1 fuzzy logic based approach is proposed. The proposed filter is simple and very effective to remove SAP noise. This filter also preserves image features such as edges, corners, etc. Moreover, it doesn’t require any parameter tuning. The proposed filter is validated on several standard grayscale images and provides improved peak signal-to-noise ratio. The paper is organized as follows. The required preliminaries for proposed methodology are presented in Section II. The proposed adaptive Type-2 fuzzy filter is explained in Section III. In Section IV, experimental results, discussion, and comparison with existing filters are presented. The concluding remarks are drawn in Section V.