Refactor gaussian_blur in numpy to utilize scipy.signal.convolve2d

[shashaka]

To improve performance, I optimized the convolution logic in gaussian_blur by replacing the NumPy-based implementation with scipy.signal.convolve2d.
Below are the performance test results on my local machine:

• Before Optimization:

%%time
%load_ext autoreload
%autoreload 2

from keras.src.backend.numpy.image import gaussian_blur

images = np.random.uniform(size=(2, 800, 800, 3)).astype("float32")
kernel_size = (3, 3)
sigma = np.random.uniform(size=(2,)).astype("float32")

transformed_image = gaussian_blur(images, sigma=sigma)

CPU times: user 7.59 s, sys: 0 ns, total: 7.59 s
Wall time: 7.59 s

• After Optimization:

%%time
%load_ext autoreload
%autoreload 2

from keras.src.backend.numpy.image import gaussian_blur

images = np.random.uniform(size=(2, 800, 800, 3)).astype("float32")
kernel_size = (3, 3)
sigma = np.random.uniform(size=(2,)).astype("float32")

transformed_image = gaussian_blur(images, sigma=sigma)

CPU times: user 123 ms, sys: 0 ns, total: 123 ms
Wall time: 122 ms

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 82.43%. Comparing base (d1fb581) to head (3d21b6c).

Additional details and impacted files

@@            Coverage Diff             @@
##           master   #20974      +/-   ##
==========================================
- Coverage   82.43%   82.43%   -0.01%     
==========================================
  Files         561      561              
  Lines       53204    53201       -3     
  Branches     8242     8241       -1     
==========================================
- Hits        43857    43854       -3     
  Misses       7338     7338              
  Partials     2009     2009              

Flag	Coverage Δ
keras	82.24% <100.00%> (-0.01%)	⬇️
keras-jax	64.03% <0.00%> (+<0.01%)	⬆️
keras-numpy	58.85% <100.00%> (-0.01%)	⬇️
keras-openvino	32.61% <0.00%> (+<0.01%)	⬆️
keras-tensorflow	64.48% <0.00%> (+<0.01%)	⬆️
keras-torch	64.09% <0.00%> (+<0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[fchollet]

LGTM, thank you