In this paper, we propose simple and effective compression of CSLBP (Center Symmetric Local Binary Pattern) descriptors, which is a textured based operator and mostly used as key point descriptor. With default parameters for computation, it is 256-length descriptor for each keypoint or affine patch. CSLBP is an extended form of LBP (Local Binary Patterns). The calculation of CSLBP descriptor is effective, robust, and straightforward for different image transformations for instance; image blurring and illumination alteration. However, an improvement in time and space consumption of CSLBP can be attained by means of simple compression. For this reason, CSLBP is a smart choice for smart phones as well as large databases. We reduce the descriptor length (dimensions) upto 50% without applying any techniques of dimensionality reduction like PCA (Principle Component Analysis) or LDA (Linear Discriminant Analysis). The compressed CSLBP descriptor is denoted as C-CSLBP. The performance of C-CSLBP is evaluated on state-of-the-art datasets using standard metrics. It is quantitatively shown by experiments that C-CSLBP is equivalently effective compared to CSLBP despite of reduced dimensions.

Sindhi language is script language like Arabic and Persian. It's origin is 2500 years old and spoken in various countries in Asia. In this paper, we propose an Optical Character Recognition (OCR) system which recognizes handwritten Sindhi numeral expressions (i.e. Sindhi handwritten numeral strings) without using common input devices such as keyboard and storage device memory. Our experiments focus on character recognition which later can be used for various applications such as tutoring, mathematical kids games, and automatic telephone number conversion from sign boards in India and Pakistan. In our research, we investigate the correlation between the numeral shapes and apply famous state-of-the art classifier based on correlation based template matching. We experimentally show that template matching gives poor performance as the shapes of numerals are highly correlated. There exists little volume of literature to address OCR on Sindhi language but unavailability of benchmark dataset makes it difficult for researchers around the world to re-implement the literature frameworks. We provide two sets of images which can be used for training and prediction.