This early-stage study presented image processing and machine learning techniques applied to images obtained during hydraulic fracture experiments. Three experiments were conducted on Opalinus Shale rock, where prismatic rock specimens (4”x2”x1”) were subjected to quasi-true triaxial stresses (σ_1=3 MPa,σ_2=2 MPa,σ_3=1 MPa) to simulate in-situ stresses and an artificial pre-cut fracture to hydraulic pressure (constant Q=20μL/s). The specimens varied in bedding plane orientation (θ=30°,45°,60°). Images taken at 1fps using a 42MP camera captured the fracturing process. Hand traced sketches were drawn for each frame to identify the location of the fracture; these sketches were then used to conduct a novel quantitatively based analysis of fracture behavior. The quantitative analysis performed on these sketches relied heavily on their spatio-temporal accuracy. An image processing algorithm was developed to analyze each sketch, segmenting the fracture networks into their individual constituents (branches), and calculate their lengths and orientations. The data were then analyzed from two lenses: static analysis incorporated all of the fractures in a sketch, and dynamic analysis incorporated only newly developed fractures in a sketch. From the static analysis, the branch lengths comprising a fracture network at any point in time were found to be exponentially distributed with a decay constant λ~1. From the dynamic analysis, the branch growth over time were found to follow a power law with slopes m=-1 to-2, providing experimental evidence that fracture growth scales proportionally. Moreover, the power law exponent of fractures growing along bedding were larger than those growing normal to bedding; the incremental growth along bedding was dominated by shorter increments. Such findings were only possible because of the new quantitative fracture analysis presented. Manually generating the sketches for analysis is a time-intensive task, so a fracture detecting machine learning algorithm was developed, trained, and implemented to autonomously generate the fracture sketches required for the quantitative analysis. A pretrained, state of the art convolutional neural network (CNN) architecture was re-trained to detect fractures in the experimental setup on a pixel-by-pixel basis. The sketches generated by the neural network were processed and used to conduct the quantitative analysis. These results were compared against those obtained from the manual sketches. The decay constants from the static analysis and the power law exponents from the dynamic analysis were used as comparative benchmarks and were in agreement with those obtained from the manual sketches. Implementing this algorithm opens up doors to analyze more experiments in a shorter time span, including past experiments conducted on various materials such as granite for EGS applications and future experiments.