This paper proposes two novel, robust versions of a local texture descriptor. Both are based on the Circular Parts Local Binary Pattern (CPLBP) approach, designed to enhance the classical Local Binary Pattern (LBP), modifying the neighborhood topology to identify many micro and large-scale textural structures. The proposal methods consider a circular neighbourhood divided into parts, calculate the median and trimmed mean of the grey values of each part, respectively, and thresholded with the grey value of the central pixel. Also, a parallel algorithm of our descriptors is presented. To put their performance into perspective, using classification accuracy measures, they are compared with the LBP and the CPLBP on two texture datasets with different characteristics: Brodatz and UIUC and under textures with increasing levels of additive white Gaussian noise (AWGN). The discriminating power of the texture descriptors was achieved using 10-fold cross-validation of a 1-Nearest Neighbor classifier and was tested with a Repeated Measures ANOVA model design. Under the uncontaminated textures, the proposed descriptors showed similar behaviour compared to the LBP and CPLBP and overall good scores at low and moderate AWGN levels. It outperforms mean accuracy at textures with high noise levels. Additionally, our methods show promising performance in recognizing green turtles in noisy images, highlighting a solid ability to distinguish textural patterns in uncontrolled environments compared to noise-robust LBP-based descriptors.