Research on bit selection in rotary drilling work and its relationship with different geological, operational parameters and bit type demands an improved procedure to select the best bit. Most researchers and studies done currently have failed to account for the bit shape and type among other features which have significance in bit selection and ROP prediction. Thus, this thesis demonstrates a pathway where bits can be easily photographed and have its features extracted in numerical form for the ROP prediction. Several objectives were put forth including extracting drill bit features from 18 different bit images, each with a unique IADC bit code in combination with drilling parameters from previous bit records to effectively create a model to predict the ROP in order to create a new method for bit selection. After pre-processing the images, 13 unique features were selected; area, perimeter, major and minor axis length, eccentricity, equivalent diameter, convex area, orientation, contrast, homogeneity and energy which were compressed into two principal components in order to simplify the prediction stage. Using the multiple linear regression, ROP was predicted with R2 and MSE of 0.47and 3.325 respectively. Using ANN showed a large improvement in R2 and MSE values where the values were 0.08579 and 0.9347 respectively. The last method was using PSO to train the ANN for a better ROP prediction. Better R2 and MSE combinations were evident giving values of 0.9434 and 0.03918 respectively where the input data consisted of both drilling parameters and bit image features.