In this paper, we investigate the ability of the Long short term memory (LSTM) recurrent neural network architecture to perform texture classification on images. Existing approaches to texture classification rely on manually designed preprocessing steps or selected feature extractors. Since LSTM networks are able to bridge over long time lags, we propose applying them directly on the image, circumventing any handcrafted prepocessing. We investigate different approaches with several input and output representations. In our experiments on a number of widely used texture benchmarking tasks (KTH-TIPS, OuTex, VisTexL, VisTexP, and NewbarkTex), we show that the performance is comparable to, or better than, existing state-ofthe-art methods for texture classification.Texture is a rich source of information about the contents of images and identity of objects. However, reliable texture recognition has been challenging because texture is a property of image pixels that is both stochastic and non-local. Most approaches to texture recognition manually design feature extractors to cope with the non-locality, choosing specific ways of integrating information about a region that is robust to changes in phase. Examples of such an approach are Haralick's texture features [1]. Numerous approaches using texture feature extractors had been investigated in 70-80's including Haralick [2], Gabor filters [3], wavelets [4] and grey level co-occurrence matrices (GLCM) [2], [1]. The main drawbacks of these approaches are the need to select the proper size of filter bank or neighborhoods and that they are computationally expensive. These methods were also applicable only on grayscale images.More recently, Drimbarean and Whelan [5], Mäenpää and Pietikäinen [6] and Iakovidis et al. [7] have incorporated color data into texture descriptors. Their works have been focused on the combination of color and texture either jointly or separately. Different texture descriptors under various color space were compared and different ways of combination were comparatively evaluated. For instance, Discrete Cosine Transform, Gabor filters, and co-occurrence matrices under separate or combined color channel. Their experiments have shown that joint color texture descriptors improve the performance. However, it is unclear what is the best way and method to describe a wide range of textures. It has so far been lacking a general and comprehensive framework to classify textures. It needs either static condition of texture or to be manually designed to obtain an optimal solution.Only a few methods, which unify the system between the feature extraction and classification step (i.e. machine learning based methods), were proposed to overcome this problem mentioned above. First, in [8], multichannel filtering scheme is combined with the neural network. More recently, Convolutional neural network [9] and Random neural network [10] were used for texture classification. Among all neural networks based approaches, Convolutional neural network has been successfully applied ...