A-Small-Image-Processing-Library

An image processing library with basic features created in C++

The library is completed in a series of steps (subtasks), completed in order.

Subtask 1

Docutmentation
Implement the following functions using 32 bit float as datatype.

Convolution of a square input matrix and a square kernel, both matrices of any size and the kernel smaller than the input,
- with and without padding to maintain or reduce the size of the input matrix.
- implement the function both as convolution and as matrix multiplication.
Non-linear activations of an input matrix of any size with rlu (Rectified Linear Units) and tanh (Hypeborlic) functions on individual matrix elements.
Sub-sampling of square input matrices of any size with max pooling and average pooling functions.
Converting a vector of random floats to a vector of probabilities with softmax and sigmoid functions.

Docutmentation
Extend and optimise the previous design by implementing multi-threading.

Use POSIX Threads to implement various functionalities including matrix-multiplications to parrallely execute associative/independent operations.
Use other open source libraries such as mkl and openblas, to achieve the same extension to your desing.
Compare performance and plot the result. Use gnuplot for plotting.
Results : We were very well able to compete with open-blas but mkl had an advantage as it made use of architecture much better than us.

Docutmentation
Use the present design to implement LeNet architecture for digit-recognition.

Result : We achieved an accuracy of (98.9%) when testing our design on MNIST data set.
Dependencies : We have used python cv2 for converting our image to a desirable format.